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
|
package com.mbien.opencl;
import java.io.IOException;
import java.nio.ByteBuffer;
import java.util.Map;
import org.junit.BeforeClass;
import org.junit.Test;
import static org.junit.Assert.*;
import static java.lang.System.*;
import static com.mbien.opencl.TestUtils.*;
import static com.sun.gluegen.runtime.BufferFactory.*;
/**
* Test testing the high level bindings.
* @author Michael Bien
*/
public class HighLevelBindingTest {
@BeforeClass
public static void setUpClass() throws Exception {
out.println("OS: " + System.getProperty("os.name"));
out.println("VM: " + System.getProperty("java.vm.name"));
}
@Test
public void contextlessTest() {
out.println(" - - - highLevelTest; contextless - - - ");
CLPlatform[] clPlatforms = CLPlatform.listCLPlatforms();
for (CLPlatform platform : clPlatforms) {
out.println("platform info:");
out.println(" name: "+platform.getName());
out.println(" profile: "+platform.getProfile());
out.println(" version: "+platform.getVersion());
out.println(" vendor: "+platform.getVendor());
CLDevice[] clDevices = platform.listCLDevices();
for (CLDevice device : clDevices) {
out.println("device info:");
out.println(" name: "+device.getName());
out.println(" profile: "+device.getProfile());
out.println(" vendor: "+device.getVendor());
out.println(" type: "+device.getType());
out.println(" global mem: "+device.getGlobalMemSize()/(1024*1024)+" MB");
out.println(" local mem: "+device.getLocalMemSize()/1024+" KB");
out.println(" clock: "+device.getMaxClockFrequency()+" MHz");
out.println(" max work group size: "+device.getMaxWorkGroupSize());
out.println(" max compute units: "+device.getMaxComputeUnits());
out.println(" extensions: "+device.getExtensions());
}
}
}
@Test
public void vectorAddGMTest() throws IOException {
out.println(" - - - highLevelTest; global memory kernel - - - ");
CLContext context = CLContext.create();
CLDevice[] contextDevices = context.getCLDevices();
out.println("context devices:");
for (CLDevice device : contextDevices) {
out.println(" "+device.toString());
}
CLProgram program = context.createProgram(getClass().getResourceAsStream("testkernels.cl")).build();
CLDevice[] programDevices = program.getCLDevices();
assertEquals(contextDevices.length, programDevices.length);
out.println("program devices:");
for (CLDevice device : programDevices) {
out.println(" "+device.toString());
out.println(" build log: "+program.getBuildLog(device));
out.println(" build status: "+program.getBuildStatus(device));
}
String source = program.getSource();
assertFalse(source.trim().isEmpty());
// out.println("source:\n"+source);
int elementCount = 11444777; // Length of float arrays to process (odd # for illustration)
int localWorkSize = 256; // set and log Global and Local work size dimensions
int globalWorkSize = roundUp(localWorkSize, elementCount); // rounded up to the nearest multiple of the LocalWorkSize
out.println("allocateing buffers of size: "+globalWorkSize);
ByteBuffer srcA = newDirectByteBuffer(globalWorkSize*SIZEOF_INT);
ByteBuffer srcB = newDirectByteBuffer(globalWorkSize*SIZEOF_INT);
ByteBuffer dest = newDirectByteBuffer(globalWorkSize*SIZEOF_INT);
fillBuffer(srcA, 23456);
fillBuffer(srcB, 46987);
CLBuffer clBufferA = context.createBuffer(CL.CL_MEM_READ_ONLY, srcA);
CLBuffer clBufferB = context.createBuffer(CL.CL_MEM_READ_ONLY, srcB);
CLBuffer clBufferC = context.createBuffer(CL.CL_MEM_WRITE_ONLY, dest);
Map<String, CLKernel> kernels = program.getCLKernels();
for (CLKernel kernel : kernels.values()) {
out.println("kernel: "+kernel.toString());
}
assertNotNull(kernels.get("VectorAddGM"));
assertNotNull(kernels.get("Test"));
CLKernel vectorAddKernel = kernels.get("VectorAddGM");
vectorAddKernel.setArg(0, SIZEOF_LONG, clBufferA)
.setArg(1, SIZEOF_LONG, clBufferB)
.setArg(2, SIZEOF_LONG, clBufferC)
.setArg(3, SIZEOF_INT, elementCount);
CLCommandQueue queue = programDevices[0].createCommandQueue();
// Asynchronous write of data to GPU device, blocking read later
queue.putWriteBuffer(clBufferA, false)
.putWriteBuffer(clBufferB, false)
.putNDRangeKernel(vectorAddKernel, 1, null, new long[]{ globalWorkSize }, new long[]{ localWorkSize })
.putReadBuffer(clBufferC, true).release();
out.println("a+b=c result snapshot: ");
for(int i = 0; i < 10; i++)
out.print(dest.getInt()+", ");
out.println("...; "+dest.remaining()/SIZEOF_INT + " more");
assertTrue(3 == context.getCLBuffers().size());
clBufferA.release();
assertTrue(2 == context.getCLBuffers().size());
assertTrue(2 == context.getCLBuffers().size());
clBufferB.release();
assertTrue(1 == context.getCLBuffers().size());
assertTrue(1 == context.getCLBuffers().size());
clBufferC.release();
assertTrue(0 == context.getCLBuffers().size());
assertTrue(1 == context.getCLPrograms().size());
program.release();
assertTrue(0 == context.getCLPrograms().size());
// CLDevice device = ctx.getMaxFlopsDevice();
// out.println("max FLOPS device: " + device);
context.release();
}
@Test
public void writeCopyReadBufferTest() throws IOException {
out.println(" - - - highLevelTest; copy buffer test - - - ");
final int elements = 10000000; //many..
CLContext context = CLContext.create();
// the CL.MEM_* flag is probably completly irrelevant in our case since we do not use a kernel in this test
CLBuffer clBufferA = context.createBuffer(CL.CL_MEM_READ_ONLY, elements*SIZEOF_INT);
CLBuffer clBufferB = context.createBuffer(CL.CL_MEM_READ_ONLY, elements*SIZEOF_INT);
// fill only first read buffer -> we will copy the payload to the second later.
fillBuffer(clBufferA.buffer, 12345);
CLCommandQueue queue = context.getCLDevices()[0].createCommandQueue();
// asynchronous write of data to GPU device, blocking read later to get the computed results back.
queue.putWriteBuffer(clBufferA, false) // write A
.putCopyBuffer(clBufferA, clBufferB, clBufferA.buffer.capacity()) // copy A -> B
.putReadBuffer(clBufferB, true); // read B
context.release();
ByteBuffer a = clBufferA.buffer;
ByteBuffer b = clBufferB.buffer;
// print first few elements of the resulting buffer to the console.
out.println("validating computed results...");
for(int i = 0; i < elements; i++) {
int aVal = a.getInt();
int bVal = b.getInt();
if(aVal != bVal) {
out.println("a: "+aVal);
out.println("b: "+bVal);
out.println("position: "+a.position());
fail("a!=b");
}
}
out.println("results are valid");
}
}
|
