1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
|
/*
* Copyright 2021 Xilinx Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <algorithm>
#include <chrono>
#include <cstring>
#include <iostream>
#include <numeric>
#include <thread>
#include <vart/runner.hpp>
#include <vitis/ai/env_config.hpp>
#include <vitis/ai/thread_pool.hpp>
#include <xir/graph/graph.hpp>
#include <xir/tensor/tensor.hpp>
#include "common.h"
using namespace std;
#define MAX_THREADS 16
#define NUM_THREADS 2
#define CORES 2
void run(std::unique_ptr<vart::Runner> runner, const std::string test_dir,
int num_sequences, int thread_idx, int loop_for) {
auto prologue = chrono::steady_clock::now();
int batch_size = runner->get_input_tensors()[0]->get_shape().at(0);
int aligned_input_seq_dim = runner->get_input_tensors()[0]->get_shape().at(2);
int aligned_output_seq_dim =
runner->get_output_tensors()[0]->get_shape().at(2);
auto vector_bin = test_dir + "/vector_bin_batch" + std::to_string(batch_size);
auto bench_bin = test_dir + "/bench_bin_batch" + std::to_string(batch_size);
auto curr_input_tensor = xir::Tensor::create(
"iv", {batch_size, num_sequences, aligned_input_seq_dim},
xir::DataType{xir::DataType::XINT, 16});
auto curr_output_tensor = xir::Tensor::create(
"ov", {batch_size, num_sequences, aligned_output_seq_dim},
xir::DataType{xir::DataType::XINT, 16});
size_t input_size = curr_input_tensor->get_element_num();
size_t output_size = curr_output_tensor->get_element_num();
char *vector_data_, *bench_data_;
size_t vector_size_, bench_size_;
std::tie(vector_data_, vector_size_) = read_binary_file(vector_bin);
std::tie(bench_data_, bench_size_) = read_binary_file(bench_bin);
std::vector<int16_t> input_vector(input_size);
int output_batch_len = num_sequences * aligned_output_seq_dim;
int batch_len = num_sequences * aligned_input_seq_dim;
for (int b = 0; b < batch_size; b++) {
memcpy(input_vector.data() + b * batch_len, vector_data_,
batch_len * sizeof(int16_t));
}
std::vector<int16_t> output_vector(output_size, 0);
auto input_tb = std::make_unique<CpuFlatTensorBuffer>(
(void*)(input_vector.data()), std::move(curr_input_tensor.get()));
auto output_tb = std::make_unique<CpuFlatTensorBuffer>(
(void*)(output_vector.data()), std::move(curr_output_tensor.get()));
std::vector<vart::TensorBuffer*> inputsPtr{input_tb.get()};
std::vector<vart::TensorBuffer*> outputsPtr{output_tb.get()};
auto loop_start = chrono::steady_clock::now();
for (auto loop_idx = 0; loop_idx < loop_for; loop_idx++) {
runner->execute_async(inputsPtr, outputsPtr);
for (int b = 0; b < batch_size; ++b) {
auto cmpret = std::memcmp((char*)bench_data_,
output_vector.data() + b * output_batch_len,
output_batch_len * sizeof(int16_t));
if (cmpret != 0) {
LOG(ERROR) << "[ERROR] Thread: " << thread_idx
<< " | Batch: " << loop_idx << " | Sentence: " << b
<< " | compare result: " << cmpret;
int outsum = std::accumulate(
output_vector.begin() + b * output_batch_len,
output_vector.begin() + (b + 1) * output_batch_len, 0);
int16_t* bd = reinterpret_cast<int16_t*>(bench_data_);
int refsum = std::accumulate(bd, bd + output_batch_len, 0);
LOG(INFO) << "Thread: " << thread_idx << " | Batch: " << loop_idx
<< " | Sentence: " << b << " | Ref Sum: " << refsum
<< " | Out Sum: " << outsum;
}
}
}
auto loop_end = chrono::steady_clock::now();
auto time_taken =
chrono::duration<float, std::milli>{loop_end - loop_start}.count();
LOG(INFO)
<< "Thread(" << thread_idx << ") execute_async took [ms]: " << time_taken
<< " | Avg. time per batch [ms]: " << time_taken / loop_for
<< " | Init Over-head [ms]: "
<< chrono::duration<float, std::milli>{loop_start - prologue}.count();
}
int main(int argc, char* argv[]) {
CHECK(argc > 5)
<< "please input the correct parameters in console: "
<< "./test_rnn_runner <xmodel_dir> <test_dir> <num_sequences> "
<< " <num_threads> <num_batches>\n";
std::string xmodel_dir = argv[1];
std::string test_dir = argv[2];
int num_sequences = atoi(argv[3]);
int num_threads = atoi(argv[4]);
int num_batches = atoi(argv[5]);
CHECK(num_threads <= MAX_THREADS)
<< "num_threads should be less than " << MAX_THREADS;
if (num_threads == 0) num_threads = NUM_THREADS;
LOG(INFO) << "\n xmodel_dir: " << xmodel_dir << "\n test_dir: " << test_dir
<< "\n num_sequences: " << num_sequences
<< "\n num_threads: " << num_threads
<< "\n num_batches: " << num_batches;
std::vector<std::string> xmodel_files;
xmodel_files.reserve(2);
xmodel_files.push_back(xmodel_dir + "/compiled_batch_3.xmodel");
xmodel_files.push_back(xmodel_dir + "/compiled_batch_4.xmodel");
std::vector<std::unique_ptr<vart::Runner>> runners;
runners.reserve(num_threads);
for (int i = 0; i < num_threads; i++) {
std::string xmodel_file = xmodel_files.at(i % 2);
auto graph = xir::Graph::deserialize(xmodel_file);
auto rs = graph->get_root_subgraph();
LOG(INFO) << "Creating Runner(" << i << ") with " << xmodel_file;
runners.push_back(vart::Runner::create_runner(rs, "run"));
}
auto start = chrono::steady_clock::now();
array<thread, MAX_THREADS> threads_list;
for (int i = 0; i < num_threads; i++) {
threads_list[i] = thread(run, std::move(runners.at(i)), test_dir,
num_sequences, i, num_batches);
}
for (int i = 0; i < num_threads; i++) {
threads_list[i].join();
}
auto end = chrono::steady_clock::now();
auto time_taken = chrono::duration<float, std::milli>{end - start}.count();
LOG(INFO) << "Time taken by all the threads [ms]: " << time_taken;
LOG(INFO) << "Throughput (batch/sec) : "
<< num_threads * num_batches * 1000.0f / time_taken;
return 0;
}
|
