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
|
#include <iostream>
#include <limits>
#include "../driver.hpp"
#include "miopen/check_numerics.hpp"
#include "miopen/handle.hpp"
#include "../tensor_holder.hpp"
#include <miopen/convolution.hpp>
#include <gtest/gtest.h>
namespace fs = miopen::fs;
const std::string test_file_name_prefix = "dumptensortest_";
const size_t tensor_size = 20;
const size_t nan_index = 5;
template <class T>
void prettyPrintTensor(const tensor<T>& host_tensor)
{
for(std::size_t i = 0; i < host_tensor.desc.GetElementSize(); i++)
{
std::cerr << host_tensor[i] << ",";
if((i + 1) % tensor_size == 0)
std::cerr << "\n";
}
std::cerr << "\n";
}
template <class T>
void populateTensor(tensor<T>& host_tensor)
{
// batch (n) = 1, channels(c) = 1, height(h) = tensor_size, width(w) = tensor_size
host_tensor = tensor<T>{1, 1, tensor_size, tensor_size}.generate(tensor_elem_gen_integer{
100}); // populate tensor with randomly generated element from [0-100]
}
template <class T>
void populateWithNAN(std::vector<T>& data)
{
std::fill(data.begin(), data.end(), std::numeric_limits<T>::quiet_NaN());
}
template <class T>
void compare(const tensor<T>& host_tensor,
const tensor<T>& tensor_from_file,
bool compare_nan = false)
{
ASSERT_EQ(host_tensor.data.size(), tensor_from_file.data.size());
for(int i = 0; i < host_tensor.data.size(); ++i)
{
// if(compare_nan && i == nan_index)
if(compare_nan)
{
EXPECT_TRUE(std::isnan(host_tensor[i])) << "Was expecting nan at index " << i;
EXPECT_TRUE(std::isnan(tensor_from_file[i])) << "Was expecting nan at index " << i;
}
else
{
T tolerance = static_cast<T>(10);
T threshold = std::numeric_limits<T>::epsilon() * tolerance;
EXPECT_NEAR(host_tensor[i], tensor_from_file[i], threshold)
<< "Vectors host_tensor and tensor_from_file differ at index " << i;
}
}
}
template <typename T>
void readBufferFromFile(T* data, size_t dataNumItems, const std::string& fileName)
{
std::ifstream infile(fileName, std::ios::binary);
if(infile)
{
infile.read(reinterpret_cast<char*>(data), dataNumItems * sizeof(T));
infile.close();
}
else
{
EXPECT_TRUE(false) << "Could not open file : " << fileName;
}
}
template <class T>
void testDump(const std::string& test_file_name)
{
tensor<T> host_tensor;
populateTensor(host_tensor);
miopen::Handle handle{};
// copy tensor from host to gpu
auto gpu_tensor_addr = handle.Write(host_tensor.data);
miopen::DumpTensorToFileFromDevice(
handle, host_tensor.desc, gpu_tensor_addr.get(), test_file_name);
// read back tensor
tensor<T> tensor_from_file = tensor<T>{1, 1, tensor_size, tensor_size};
readBufferFromFile<T>(
tensor_from_file.data.data(), host_tensor.desc.GetElementSpace(), test_file_name);
compare(host_tensor, tensor_from_file);
// clean up
fs::remove(test_file_name);
}
template <class T>
void testDumpWithNan(const std::string& test_file_name)
{
tensor<T> host_tensor;
populateTensor(host_tensor);
miopen::Handle handle{};
// before writing to gpu we set one of the element
// in the vector to nan.
// host_tensor.data[nan_index] = std::numeric_limits<T>::quiet_NaN();
populateWithNAN(host_tensor.data);
// write the tensor to GPU
auto gpu_tensor_addr = handle.Write(host_tensor.data);
miopen::DumpTensorToFileFromDevice(
handle, host_tensor.desc, gpu_tensor_addr.get(), test_file_name);
if(miopen::checkNumericsInput(handle, host_tensor.desc, gpu_tensor_addr.get()))
{
// read back tensor
tensor<T> tensor_from_file = tensor<T>{1, 1, tensor_size, tensor_size};
readBufferFromFile<T>(
tensor_from_file.data.data(), host_tensor.desc.GetElementSpace(), test_file_name);
compare(host_tensor, tensor_from_file, true);
}
else
{
EXPECT_TRUE(false)
<< "Was expecting NAN value in tensor. Current value at host_tensor.data[" << nan_index
<< "] = " << host_tensor.data[nan_index];
}
// clean up
fs::remove(test_file_name);
}
namespace {
std::string GetFullFileName(const std::string& filename)
{
return fs::temp_directory_path().append(test_file_name_prefix + filename).string();
}
} // namespace
TEST(CPU_Dump_FP32, testDump) { testDump<float>(GetFullFileName("float.bin")); }
TEST(CPU_Dump_NAN_FP32, testDump) { testDumpWithNan<float>(GetFullFileName("nan_float.bin")); }
TEST(CPU_Dump_FP16, testDump) { testDump<half_float::half>(GetFullFileName("half_float.bin")); }
TEST(CPU_Dump_NAN_FP16, testDump)
{
testDumpWithNan<half_float::half>(GetFullFileName("nan_half_float.bin"));
}
TEST(CPU_Dump_BFP16, testDump) { testDump<bfloat16>(GetFullFileName("bfloat16.bin")); }
TEST(CPU_Dump_NAN_BFP16, testDump)
{
testDumpWithNan<bfloat16>(GetFullFileName("nan_bfloat16.bin"));
}
|
