1 /*
2     Copyright (c) 2005-2021 Intel Corporation
3 
4     Licensed under the Apache License, Version 2.0 (the "License");
5     you may not use this file except in compliance with the License.
6     You may obtain a copy of the License at
7 
8         http://www.apache.org/licenses/LICENSE-2.0
9 
10     Unless required by applicable law or agreed to in writing, software
11     distributed under the License is distributed on an "AS IS" BASIS,
12     WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13     See the License for the specific language governing permissions and
14     limitations under the License.
15 */
16 
17 #include "common/parallel_reduce_common.h"
18 #include "common/concurrency_tracker.h"
19 
20 #include "../tbb/test_partitioner.h"
21 
22 //! \file conformance_parallel_reduce.cpp
23 //! \brief Test for [algorithms.parallel_reduce algorithms.parallel_deterministic_reduce] specification
24 
25 class RotOp {
26 public:
27     using Type = int;
28     int operator() ( int x, int i ) const {
29         return ( x<<1 ) ^ i;
30     }
31     int join( int x, int y ) const {
32         return operator()( x, y );
33     }
34 };
35 
36 template <class Op>
37 struct ReduceBody {
38     using result_type = typename Op::Type;
39     result_type my_value;
40 
41     ReduceBody() : my_value() {}
42     ReduceBody( ReduceBody &, oneapi::tbb::split ) : my_value() {}
43 
44     void operator() ( const oneapi::tbb::blocked_range<int>& r ) {
45         utils::ConcurrencyTracker ct;
46         for ( int i = r.begin(); i != r.end(); ++i ) {
47             Op op;
48             my_value = op(my_value, i);
49         }
50     }
51 
52     void join( const ReduceBody& y ) {
53         Op op;
54         my_value = op.join(my_value, y.my_value);
55     }
56 };
57 
58 template <class Partitioner>
59 void TestDeterministicReductionFor() {
60     const int N = 1000;
61     const oneapi::tbb::blocked_range<int> range(0, N);
62     using BodyType = ReduceBody<RotOp>;
63     using Type = RotOp::Type;
64 
65     BodyType benchmark_body;
66     deterministic_reduce_invoker(range, benchmark_body, Partitioner());
67     for ( int i=0; i<100; ++i ) {
68         BodyType measurement_body;
69         deterministic_reduce_invoker(range, measurement_body, Partitioner());
70         REQUIRE_MESSAGE( benchmark_body.my_value == measurement_body.my_value,
71         "parallel_deterministic_reduce behaves differently from run to run" );
72 
73         Type lambda_measurement_result = deterministic_reduce_invoker<Type>( range,
74             [](const oneapi::tbb::blocked_range<int>& br, Type value) -> Type {
75                 utils::ConcurrencyTracker ct;
76                 for ( int ii = br.begin(); ii != br.end(); ++ii ) {
77                     RotOp op;
78                     value = op(value, ii);
79                 }
80                 return value;
81             },
82             [](const Type& v1, const Type& v2) -> Type {
83                 RotOp op;
84                 return op.join(v1,v2);
85             },
86             Partitioner()
87         );
88         REQUIRE_MESSAGE( benchmark_body.my_value == lambda_measurement_result,
89             "lambda-based parallel_deterministic_reduce behaves differently from run to run" );
90     }
91 }
92 
93 //! Test that deterministic reduction returns the same result during several measurements
94 //! \brief \ref requirement \ref interface
95 TEST_CASE("Test deterministic reduce correctness") {
96     for ( auto concurrency_level : utils::concurrency_range() ) {
97         oneapi::tbb::global_control control(oneapi::tbb::global_control::max_allowed_parallelism, concurrency_level);
98         TestDeterministicReductionFor<oneapi::tbb::simple_partitioner>();
99         TestDeterministicReductionFor<oneapi::tbb::static_partitioner>();
100         TestDeterministicReductionFor<utils_default_partitioner>();
101     }
102 }
103 
104 //! Test partitioners interaction with various ranges
105 //! \brief \ref requirement \ref interface
106 TEST_CASE("Test partitioners interaction with various ranges") {
107     using namespace test_partitioner_utils::interaction_with_range_and_partitioner;
108     for ( auto concurrency_level : utils::concurrency_range() ) {
109         oneapi::tbb::global_control control(oneapi::tbb::global_control::max_allowed_parallelism, concurrency_level);
110 
111         test_partitioner_utils::SimpleReduceBody body;
112         oneapi::tbb::affinity_partitioner ap;
113 
114         parallel_reduce(Range1(/*assert_in_split*/ true, /*assert_in_proportional_split*/ false), body, ap);
115         parallel_reduce(Range6(false, true), body, ap);
116 
117         parallel_reduce(Range1(/*assert_in_split*/ true, /*assert_in_proportional_split*/ false), body, oneapi::tbb::static_partitioner());
118         parallel_reduce(Range6(false, true), body, oneapi::tbb::static_partitioner());
119 
120         parallel_reduce(Range1(/*assert_in_split*/ false, /*assert_in_proportional_split*/ true), body, oneapi::tbb::simple_partitioner());
121         parallel_reduce(Range6(false, true), body, oneapi::tbb::simple_partitioner());
122 
123         parallel_reduce(Range1(/*assert_in_split*/ false, /*assert_in_proportional_split*/ true), body, oneapi::tbb::auto_partitioner());
124         parallel_reduce(Range6(false, true), body, oneapi::tbb::auto_partitioner());
125 
126         parallel_deterministic_reduce(Range1(/*assert_in_split*/true, /*assert_in_proportional_split*/ false), body, oneapi::tbb::static_partitioner());
127         parallel_deterministic_reduce(Range6(false, true), body, oneapi::tbb::static_partitioner());
128 
129         parallel_deterministic_reduce(Range1(/*assert_in_split*/false, /*assert_in_proportional_split*/ true), body, oneapi::tbb::simple_partitioner());
130         parallel_deterministic_reduce(Range6(false, true), body, oneapi::tbb::simple_partitioner());
131     }
132 }
133