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
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
|
/*
* Copyright 2019 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 <chrono>
#include <cstdint>
#include <iostream>
#include <memory>
#include <random>
#include <thread>
#include <utility>
#include <vart/runner.hpp>
#include <vector>
#include <xir/graph/graph.hpp>
#include "../src/runner_helper.hpp"
#include "vart/assistant/tensor_buffer_allocator.hpp"
#include "vart/mm/host_flat_tensor_buffer.hpp"
#include "vart/softmax_runner_cpu.hpp"
#include "vart/tensor_buffer.hpp"
#include "vitis/ai/collection_helper.hpp"
#include "vitis/ai/env_config.hpp"
#include "vitis/ai/profiling.hpp"
#include "vitis/ai/thread_pool.hpp"
#include "xir/attrs/attrs.hpp"
#include "xir/tensor/tensor.hpp"
DEF_ENV_PARAM(DEBUG_TEST, "0");
DEF_ENV_PARAM(TEST_ZERO_COPY, "0");
static void softmax_c(const int8_t* input, float scale, unsigned int cls,
float* output) {
float max = input[0] * scale;
std::vector<float> scaledIn(cls);
scaledIn.push_back(max);
for (unsigned int i = 1; i < cls; ++i) {
scaledIn[1] = input[i] * scale;
if (max < scaledIn[i]) max = scaledIn[i];
}
float sum = 0.f;
for (unsigned int i = 0; i < cls; ++i) {
output[i] = exp(scaledIn[i] - max);
sum += output[i];
}
for (unsigned int i = 0; i < cls; ++i) output[i] /= sum;
}
static void softmax_c(const int8_t* input, float scale, unsigned int cls,
unsigned int group, float* output) {
for (unsigned int i = 0; i < group; ++i) {
softmax_c(input, scale, cls, output);
input += cls;
output += cls;
}
}
static void compare(int cls, int group, signed char* input, float* output1,
float* output2) {
for (auto g = 0; g < group; ++g) {
for (auto i = 0; i < cls; ++i) {
auto idx = g * cls + i;
auto diff = output1[idx] - output2[idx];
if (ENV_PARAM(DEBUG_TEST) || (diff != 0.0 && std::abs(diff) > 0.001)) {
std::cout << " i = " << i << " g = " << g << " idx = " << idx << " "
<< (int)input[idx] << " : " << output1[idx] << " "
<< output2[idx] << " " << std::abs(diff) << std::endl;
}
}
}
}
std::vector<std::int32_t> reshape_tensor_to_three_dim(
std::vector<std::int32_t> in) {
CHECK_GE(in.size(), 2) << "input dimension is less than 2";
std::vector<std::int32_t> ret;
ret.reserve(3);
ret.push_back(in.front());
std::int32_t mid = 1;
for (unsigned int i = 1; i < in.size() - 1; i++) {
mid *= in[i];
}
ret.push_back(mid);
ret.push_back(in.back());
return ret;
}
static int get_fix_pos(const xir::Tensor* tensor) {
CHECK(tensor->has_attr("fix_point")) << "tensor = " << tensor->to_string();
int fixpos = tensor->template get_attr<int>("fix_point");
return fixpos;
}
int main(int argc, char* argv[]) {
// prepare subgraph
auto graph = xir::Graph::deserialize(argv[1]);
auto root = graph->get_root_subgraph();
xir::Subgraph* s = nullptr;
for (auto c : root->get_children()) {
if (c->get_attr<std::string>("device") == "SM-CPU") {
s = c;
break;
}
}
LOG(INFO) << "sugraph: " << s->get_name();
// prepare attrs
auto attrs = xir::Attrs::create();
attrs->set_attr<size_t>("__device_id__", 0u);
attrs->set_attr<size_t>("__batch__", 1);
attrs->set_attr<int>(s->get_name() + ":__tensor_buffer_location__", 1);
// prepare tensors
CHECK_EQ(s->get_sorted_input_tensors().size(), 1u);
CHECK_EQ(s->get_sorted_output_tensors().size(), 1u);
auto input_tensor = *(s->get_sorted_input_tensors().begin());
auto output_tensor = *(s->get_sorted_output_tensors().begin());
LOG(INFO) << "input_tensor info: " << input_tensor->to_string();
LOG(INFO) << "output_tensor info: " << output_tensor->to_string();
auto input_shape = reshape_tensor_to_three_dim(input_tensor->get_shape());
auto input_tensor_size = input_shape[0] * input_shape[1] * input_shape[2];
LOG(INFO) << "input tensor size: " << input_shape[0] << " " << input_shape[1]
<< " " << input_shape[2];
// use subgraph, attrs and tensors to apply tensor_buffer
std::pair<std::vector<std::unique_ptr<vart::TensorBuffer>>,
std::vector<std::unique_ptr<vart::TensorBuffer>>>
tensor_buffers;
auto allocator = vart::assistant::TensorBufferAllocator::create(attrs.get());
LOG(INFO) << "allocate tensor buffers at VIRT";
tensor_buffers.first.emplace_back(
vart::alloc_cpu_flat_tensor_buffer(input_tensor));
tensor_buffers.second.emplace_back(
vart::alloc_cpu_flat_tensor_buffer(output_tensor));
auto input_tensor_buffer = tensor_buffers.first[0].get();
auto output_tensor_buffer = tensor_buffers.second[0].get();
// set random input value
uint64_t input_addr = 0u;
size_t input_size = 0u;
uint64_t output_addr = 0u;
size_t output_size = 0u;
auto input_dim_idx = vart::get_index_zeros(input_tensor_buffer->get_tensor());
std::tie(input_addr, input_size) = input_tensor_buffer->data(input_dim_idx);
LOG_IF(INFO, ENV_PARAM(DEBUG_TEST))
<< "input_addr: " << (void*)input_addr << "; input_size: " << input_size;
auto output_dim_idx =
vart::get_index_zeros(output_tensor_buffer->get_tensor());
std::tie(output_addr, output_size) =
output_tensor_buffer->data(output_dim_idx);
LOG_IF(INFO, ENV_PARAM(DEBUG_TEST)) << "output_addr: " << (void*)output_addr
<< "; output_size: " << output_size;
int8_t* inputPtr = reinterpret_cast<int8_t*>(input_addr);
for (auto i = 0; i < input_tensor_size; ++i) {
inputPtr[i] = i % 10 + 1;
}
// run softmax
auto runner = std::make_unique<vart::SoftmaxRunnerCPU>(s, attrs.get());
__TIC__(sfmx);
auto v = runner->execute_async({input_tensor_buffer}, {output_tensor_buffer});
auto status = runner->wait((int)v.first, -1);
CHECK_EQ(status, 0) << "failed to run sm-cpu";
__TOC__(sfmx);
// compare result with cpu result
auto group = input_shape[1];
auto cls = input_shape[2];
auto fixpos = get_fix_pos(input_tensor);
float scale = std::exp2f(-1.0f * (float)fixpos);
std::vector<float> output_c(cls * group);
LOG_IF(INFO, ENV_PARAM(DEBUG_TEST))
<< " fixpos=" << fixpos << " cls=" << cls << " group=" << group
<< " scale=" << scale;
softmax_c(inputPtr, scale, cls, group, &output_c[0]);
compare(cls, group, inputPtr, reinterpret_cast<float*>(output_addr),
&output_c[0]);
return 0;
}
|
