1 // Copyright 2016 The Gemmlowp Authors. All Rights Reserved.
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
6 //
7 //     http://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
14 
15 #include <unistd.h>
16 #ifdef __APPLE__
17 #include <sys/time.h>
18 #endif
19 
20 #include <cstdint>
21 #include <cstdlib>
22 #include <ctime>
23 #include <iomanip>
24 #include <iostream>
25 #include <map>
26 #include <memory>
27 #include <vector>
28 
29 #include "streams.h"
30 
31 #define MUL_OFFSET (3)
32 #define ADD_OFFSET (100)
33 
34 using namespace gemmlowp::meta;
35 
36 void prepare_row_major_data(int rows, int elements, int stride, std::uint8_t* data) {
37   for (int i = 0; i < rows * stride; ++i) {
38     data[i] = 255;
39   }
40   for (int i = 0; i < rows; ++i) {
41     for (int j = 0; j < elements; ++j) {
42       data[i * stride + j] = j % 256;
43     }
44   }
45 }
46 
47 void prepare_column_major_data(int columns, int elements, int stride,
48                                std::uint8_t* data) {
49   for (int i = 0; i < elements * stride; ++i) {
50     data[i] = 255;
51   }
52   for (int i = 0; i < elements; ++i) {
53     for (int j = 0; j < columns; ++j) {
54       data[i * stride + j] = i % 256;
55     }
56   }
57 }
58 
59 void print_out(std::uint8_t* result, int rows, int elements) {
60   int size = rows * ((elements + 7) / 8) * 8;
61   for (int i = 0; i < size; ++i) {
62     std::cout << static_cast<int>(result[i]) << " ";
63   }
64   std::cout << std::endl << std::flush;
65 }
66 
67 bool check(std::uint8_t* result, int rows, int elements) {
68   int chunks = elements / 8;
69   int leftover = elements % 8;
70   for (int i = 0; i < chunks; ++i) {
71     int chunk_index = i * rows * 8;
72     int chunk_start_value = i * 8;
73     for (int j = 0; j < rows; ++j) {
74       for (int k = 0; k < 8; ++k) {
75         if (result[chunk_index + j * 8 + k] != chunk_start_value + k) {
76           return false;
77         }
78       }
79     }
80   }
81 
82   int leftover_index = chunks * rows * 8;
83   int leftover_start_value = chunks * 8;
84   for (int i = 0; i < rows; ++i) {
85     for (int j = 0; j < leftover; ++j) {
86       if (result[leftover_index + i * 8 + j] != leftover_start_value + j) {
87         return false;
88       }
89     }
90   }
91 
92   int expected_sum =
93       ((elements * (elements - 1)) / 2) * MUL_OFFSET + ADD_OFFSET;
94   int sums_offset = rows * ((elements + 7) / 8) * 8;
95   std::int32_t* sums = reinterpret_cast<std::int32_t*>(result + sums_offset);
96   for (int i = 0; i < rows; ++i) {
97     if (sums[i] != expected_sum) {
98       return false;
99     }
100   }
101 
102   return true;
103 }
104 
105 template <int lanes, int leftover>
106 void test_2(std::uint8_t* in, std::uint8_t* out) {
107   for (int elements = 8; elements < 64; elements += 8) {
108     int all_elements = elements + leftover;
109     for (int stride = all_elements; stride < all_elements + 4; ++stride) {
110       RowMajorWithSum params;
111       params.count = all_elements;
112       params.stride = stride;
113       params.multiplicative_sum_offset = MUL_OFFSET;
114       params.additive_sum_offset = ADD_OFFSET;
115 
116       prepare_row_major_data(lanes, all_elements, stride, in);
117       Stream<std::uint8_t, lanes, 8, leftover, RowMajorWithSum>::Pack(in, params,
118                                                                  out);
119       if (check(out, lanes, all_elements)) {
120         //        std::cout << "Row: " << lanes << "x8x" << leftover << " : "
121         //                  << all_elements << "@" << stride << " -- OK" <<
122         //                  std::endl;
123       } else {
124         std::cout << "Row: " << lanes << "x8x" << leftover << " : "
125                   << all_elements << "@" << stride << " -- ERROR" << std::endl;
126         std::cout << "Exiting." << std::endl;
127         std::exit(1);
128       }
129     }
130 
131     for (int stride = lanes; stride < lanes + 4; ++stride) {
132       ColumnMajorWithSum params;
133       params.count = all_elements;
134       params.stride = stride;
135       params.multiplicative_sum_offset = MUL_OFFSET;
136       params.additive_sum_offset = ADD_OFFSET;
137 
138       prepare_column_major_data(lanes, all_elements, stride, in);
139       Stream<std::uint8_t, lanes, 8, leftover, ColumnMajorWithSum>::Pack(in, params,
140                                                                     out);
141       if (check(out, lanes, all_elements)) {
142         //        std::cout << "Column: " << lanes << "x8x" << leftover << " : "
143         //                  << all_elements << "@" << stride << " -- OK" <<
144         //                  std::endl;
145       } else {
146         std::cout << "Column: " << lanes << "x8x" << leftover << " : "
147                   << all_elements << "@" << stride << " -- ERROR" << std::endl;
148         std::cout << "Exiting." << std::endl;
149         std::exit(1);
150       }
151     }
152   }
153 }
154 
155 template <int lanes>
156 void test(std::uint8_t* in, std::uint8_t* out) {
157   test_2<lanes, 0>(in, out);
158   test_2<lanes, 1>(in, out);
159   test_2<lanes, 2>(in, out);
160   test_2<lanes, 3>(in, out);
161   test_2<lanes, 4>(in, out);
162   test_2<lanes, 5>(in, out);
163   test_2<lanes, 6>(in, out);
164   test_2<lanes, 7>(in, out);
165 }
166 
167 int main() {
168   std::unique_ptr<std::uint8_t> in(new std::uint8_t[128 * 1024]);
169   std::unique_ptr<std::uint8_t> out(new std::uint8_t[128 * 1024]);
170 
171   test<1>(in.get(), out.get());
172   test<2>(in.get(), out.get());
173   test<3>(in.get(), out.get());
174   test<4>(in.get(), out.get());
175   test<5>(in.get(), out.get());
176   test<6>(in.get(), out.get());
177   test<7>(in.get(), out.get());
178   test<8>(in.get(), out.get());
179 
180   std::cout << "Ok." << std::endl;
181   return 0;
182 }
183