-
Notifications
You must be signed in to change notification settings - Fork 0
/
DepthwiseConv2D.cc
99 lines (79 loc) · 3.21 KB
/
DepthwiseConv2D.cc
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
#include <iostream>
#include <algorithm> // for copy
#include <iterator> // for ostream_iterator
#include <vector>
#include <numeric>
#include "common/nputensor.h"
int main() {
// Init
ACL_CALL(aclInit(nullptr));
ACL_CALL(aclrtSetDevice(0));
// Get Run Mode - ACL_HOST
aclrtRunMode runMode;
ACL_CALL(aclrtGetRunMode(&runMode));
std::string run_mode_str = (runMode == ACL_DEVICE) ? "ACL_DEVICE" : "ACL_HOST";
std::cout << "aclrtRunMode is : " << run_mode_str << std::endl;
// op type
const std::string op_type = "DepthwiseConv2D";
// input 0 - x
const std::vector<int64_t> input_0_dims{6, 32, 150, 150};
const std::vector<float> input_0_data(4320000, 1);
// input 1 - filter
const std::vector<int64_t> input_1_dims{1, 32, 3, 3};
const std::vector<float> input_1_data(288, 1);
// output - y
const std::vector<int64_t> output_dims{6, 32, 150, 150};
// attrs
const std::vector<int64_t> strides(4, 1);
const std::vector<int64_t> dilations(4, 1);
const std::vector<int64_t> pads(4, 1);
const std::string data_format = "NCHW";
// input tensor 0 - x
auto input_0 = new npuTensor<float>(ACL_FLOAT, input_0_dims.size(), input_0_dims.data(), ACL_FORMAT_NCHW, input_0_data.data(), memType::DEVICE);
// input tensor 1 - filter
auto input_1 = new npuTensor<float>(ACL_FLOAT, input_1_dims.size(), input_1_dims.data(), ACL_FORMAT_NCHW, input_1_data.data(), memType::DEVICE);
// set inputs desc and buffer
std::vector<aclTensorDesc *> input_descs;
std::vector<aclDataBuffer *> input_buffers;
input_descs.emplace_back(input_0->desc);
input_descs.emplace_back(input_1->desc);
input_buffers.emplace_back(input_0->buffer);
input_buffers.emplace_back(input_1->buffer);
// output - out
auto output = new npuTensor<float>(ACL_FLOAT, output_dims.size(), output_dims.data(), ACL_FORMAT_NCHW, nullptr);
// set output desc and buffer
std::vector<aclTensorDesc *> output_descs;
std::vector<aclDataBuffer *> output_buffers;
output_descs.emplace_back(output->desc);
output_buffers.emplace_back(output->buffer);
// attr
auto attr = aclopCreateAttr();
ACL_CALL(aclopSetAttrListInt(attr, "strides", strides.size(), strides.data()));
ACL_CALL(aclopSetAttrListInt(attr, "dilations", dilations.size(), dilations.data()));
ACL_CALL(aclopSetAttrListInt(attr, "pads", pads.size(), pads.data()));
ACL_CALL(aclopSetAttrString(attr, "data_format", data_format.c_str()));
// create stream
aclrtStream stream = nullptr;
ACL_CALL(aclrtCreateStream(&stream));
std::cout << "aclopCompileAndExecute : " << op_type << std::endl;
ACL_CALL(aclopCompileAndExecute(op_type.c_str(),
input_descs.size(), input_descs.data(), input_buffers.data(),
output_descs.size(), output_descs.data(), output_buffers.data(),
attr, ACL_ENGINE_SYS, ACL_COMPILE_SYS, NULL, stream));
// sync and destroy stream
ACL_CALL(aclrtSynchronizeStream(stream));
ACL_CALL(aclrtDestroyStream(stream));
// print output
input_0->Print("x");
input_1->Print("filter");
output->Print("y");
// destroy - outputs
input_0->Destroy();
input_1->Destroy();
output->Destroy();
aclopDestroyAttr(attr);
// release
ACL_CALL(aclrtResetDevice(0));
ACL_CALL(aclFinalize());
return 0;
}