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
208
209
210
|
/*
* GridTools
*
* Copyright (c) 2014-2023, ETH Zurich
* All rights reserved.
*
* Please, refer to the LICENSE file in the root directory.
* SPDX-License-Identifier: BSD-3-Clause
*/
#include "gtest/gtest.h"
#include <cstdlib>
#include <gridtools/common/atomic_functions.hpp>
#include <gridtools/common/cuda_util.hpp>
#include <gridtools/common/defs.hpp>
template <typename T>
struct verifier {
static void TestEQ(T val, T exp) {
T err = std::fabs(val - exp) / std::fabs(val);
ASSERT_TRUE(err < 1e-12);
}
};
template <>
struct verifier<float> {
static void TestEQ(float val, float exp) {
double err = std::fabs(val - exp) / std::fabs(val);
ASSERT_TRUE(err < 1e-6);
}
};
template <>
struct verifier<int> {
static void TestEQ(int val, int exp) { ASSERT_EQ(val, exp); }
};
template <typename T>
__global__ void atomic_add_kernel(T *pReduced, const T *field, const int size) {
const int i = threadIdx.x + blockIdx.x * blockDim.x;
const int j = threadIdx.y + blockIdx.y * blockDim.y;
const int pos = j * gridDim.x * blockDim.x + i;
gridtools::atomic_add(*pReduced, field[pos]);
}
template <typename T>
__global__ void atomic_sub_kernel(T *pReduced, const T *field, const int size) {
const int i = threadIdx.x + blockIdx.x * blockDim.x;
const int j = threadIdx.y + blockIdx.y * blockDim.y;
const int pos = j * gridDim.x * blockDim.x + i;
gridtools::atomic_sub(*pReduced, field[pos]);
}
template <typename T>
__global__ void atomic_min_kernel(T *pReduced, const T *field, const int size) {
const int i = threadIdx.x + blockIdx.x * blockDim.x;
const int j = threadIdx.y + blockIdx.y * blockDim.y;
const int pos = j * gridDim.x * blockDim.x + i;
gridtools::atomic_min(*pReduced, field[pos]);
}
template <typename T>
__global__ void atomic_max_kernel(T *pReduced, const T *field, const int size) {
const int i = threadIdx.x + blockIdx.x * blockDim.x;
const int j = threadIdx.y + blockIdx.y * blockDim.y;
const int pos = j * gridDim.x * blockDim.x + i;
gridtools::atomic_max(*pReduced, field[pos]);
}
template <typename T>
void test_atomic_add() {
dim3 threadsPerBlock(4, 4);
dim3 numberOfBlocks(4, 4);
int size = threadsPerBlock.x * threadsPerBlock.y * numberOfBlocks.x * numberOfBlocks.y;
T field[size];
T sumRef = 0;
T sum = 0;
T *sumDevice;
GT_CUDA_CHECK(cudaMalloc(&sumDevice, sizeof(T)));
GT_CUDA_CHECK(cudaMemcpy(sumDevice, &sum, sizeof(T), cudaMemcpyHostToDevice));
T *fieldDevice;
GT_CUDA_CHECK(cudaMalloc(&fieldDevice, sizeof(T) * size));
for (int cnt = 0; cnt < size; ++cnt) {
field[cnt] = static_cast<T>(std::rand() % 100 + (std::rand() % 100) * 0.005);
sumRef += field[cnt];
}
GT_CUDA_CHECK(cudaMemcpy(fieldDevice, &field[0], sizeof(T) * size, cudaMemcpyHostToDevice));
atomic_add_kernel<<<numberOfBlocks, threadsPerBlock>>>(sumDevice, fieldDevice, size);
GT_CUDA_CHECK(cudaMemcpy(&sum, sumDevice, sizeof(T), cudaMemcpyDeviceToHost));
verifier<T>::TestEQ(sumRef, sum);
}
template <typename T>
void test_atomic_sub() {
dim3 threadsPerBlock(4, 4);
dim3 numberOfBlocks(4, 4);
int size = threadsPerBlock.x * threadsPerBlock.y * numberOfBlocks.x * numberOfBlocks.y;
T field[size];
T sumRef = 0;
T sum = 0;
T *sumDevice;
GT_CUDA_CHECK(cudaMalloc(&sumDevice, sizeof(T)));
GT_CUDA_CHECK(cudaMemcpy(sumDevice, &sum, sizeof(T), cudaMemcpyHostToDevice));
T *fieldDevice;
GT_CUDA_CHECK(cudaMalloc(&fieldDevice, sizeof(T) * size));
for (int cnt = 0; cnt < size; ++cnt) {
field[cnt] = static_cast<T>(std::rand() % 100 + (std::rand() % 100) * 0.005);
sumRef -= field[cnt];
}
GT_CUDA_CHECK(cudaMemcpy(fieldDevice, &field[0], sizeof(T) * size, cudaMemcpyHostToDevice));
atomic_sub_kernel<<<numberOfBlocks, threadsPerBlock>>>(sumDevice, fieldDevice, size);
GT_CUDA_CHECK(cudaMemcpy(&sum, sumDevice, sizeof(T), cudaMemcpyDeviceToHost));
verifier<T>::TestEQ(sumRef, sum);
}
template <typename T>
void test_atomic_min() {
dim3 threadsPerBlock(4, 4);
dim3 numberOfBlocks(4, 4);
int size = threadsPerBlock.x * threadsPerBlock.y * numberOfBlocks.x * numberOfBlocks.y;
T field[size];
T minRef = 99999;
T min = 99999;
T *minDevice;
GT_CUDA_CHECK(cudaMalloc(&minDevice, sizeof(T)));
GT_CUDA_CHECK(cudaMemcpy(minDevice, &min, sizeof(T), cudaMemcpyHostToDevice));
T *fieldDevice;
GT_CUDA_CHECK(cudaMalloc(&fieldDevice, sizeof(T) * size));
for (int cnt = 0; cnt < size; ++cnt) {
field[cnt] = static_cast<T>(std::rand() % 100 + (std::rand() % 100) * 0.005);
minRef = std::min(minRef, field[cnt]);
}
GT_CUDA_CHECK(cudaMemcpy(fieldDevice, &field[0], sizeof(T) * size, cudaMemcpyHostToDevice));
atomic_min_kernel<<<numberOfBlocks, threadsPerBlock>>>(minDevice, fieldDevice, size);
GT_CUDA_CHECK(cudaMemcpy(&min, minDevice, sizeof(T), cudaMemcpyDeviceToHost));
verifier<T>::TestEQ(minRef, min);
}
template <typename T>
void test_atomic_max() {
dim3 threadsPerBlock(4, 4);
dim3 numberOfBlocks(4, 4);
int size = threadsPerBlock.x * threadsPerBlock.y * numberOfBlocks.x * numberOfBlocks.y;
T field[size];
T maxRef = -1;
T max = -1;
T *maxDevice;
GT_CUDA_CHECK(cudaMalloc(&maxDevice, sizeof(T)));
GT_CUDA_CHECK(cudaMemcpy(maxDevice, &max, sizeof(T), cudaMemcpyHostToDevice));
T *fieldDevice;
GT_CUDA_CHECK(cudaMalloc(&fieldDevice, sizeof(T) * size));
for (int cnt = 0; cnt < size; ++cnt) {
field[cnt] = static_cast<T>(std::rand() % 100 + (std::rand() % 100) * 0.005);
maxRef = std::max(maxRef, field[cnt]);
}
GT_CUDA_CHECK(cudaMemcpy(fieldDevice, &field[0], sizeof(T) * size, cudaMemcpyHostToDevice));
atomic_max_kernel<<<numberOfBlocks, threadsPerBlock>>>(maxDevice, fieldDevice, size);
GT_CUDA_CHECK(cudaMemcpy(&max, maxDevice, sizeof(T), cudaMemcpyDeviceToHost));
verifier<T>::TestEQ(maxRef, max);
}
TEST(AtomicFunctionsUnittest, atomic_add_int) { test_atomic_add<int>(); }
TEST(AtomicFunctionsUnittest, atomic_add_real) {
test_atomic_add<double>();
test_atomic_add<float>();
}
TEST(AtomicFunctionsUnittest, atomic_sub_int) { test_atomic_sub<int>(); }
TEST(AtomicFunctionsUnittest, atomic_sub_real) {
test_atomic_sub<double>();
test_atomic_sub<float>();
}
TEST(AtomicFunctionsUnittest, atomic_min_int) { test_atomic_min<int>(); }
TEST(AtomicFunctionsUnittest, atomic_min_real) {
test_atomic_min<double>();
test_atomic_min<float>();
}
TEST(AtomicFunctionsUnittest, atomic_max_int) { test_atomic_max<int>(); }
TEST(AtomicFunctionsUnittest, atomic_max_real) {
test_atomic_max<double>();
test_atomic_max<float>();
}
|
