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
|
// Copyright 2018 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
/* This tutorial demonstrates how MPI-parallel applications can be
* written which still render replicated data using the path tracer
* for secondary effects. Applications can leverage MPI parallelism for
* faster file I/O and load times when compared to the offload device.
* However, such applications must be written to be aware of MPI while
* the offload device behaves just like local rendering from the application's
* perspective.
*/
#include <imgui.h>
#include <mpi.h>
#include <iterator>
#include <memory>
#include <random>
#include "GLFWDistribOSPRayWindow.h"
#include "ospray/ospray_cpp.h"
#include "ospray/ospray_cpp/ext/rkcommon.h"
#include "ospray/ospray_util.h"
#include "ospray_testing.h"
#include "rkcommon/utility/getEnvVar.h"
using namespace ospray;
using namespace rkcommon;
using namespace rkcommon::math;
static std::string rendererType = "pathtracer";
static std::string builderType = "boxes";
void printHelp()
{
std::cout <<
R"description(
usage: ./ospMPIDistribTutorialReplicated [-h | --help] [[-s | --scene] scene] [[r | --renderer] renderer_type]
scenes:
boxes
cornell_box
curves
cylinders
empty
gravity_spheres_volume
perlin_noise_volumes
random_spheres
streamlines
subdivision_cube
unstructured_volume
)description";
}
int main(int argc, char **argv)
{
for (int i = 1; i < argc; ++i) {
std::string arg = argv[i];
if (arg == "-h" || arg == "--help") {
printHelp();
return 0;
} else if (arg == "-r" || arg == "--renderer") {
rendererType = argv[++i];
} else if (arg == "-s" || arg == "--scene") {
builderType = argv[++i];
}
}
int mpiThreadCapability = 0;
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &mpiThreadCapability);
if (mpiThreadCapability != MPI_THREAD_MULTIPLE
&& mpiThreadCapability != MPI_THREAD_SERIALIZED) {
fprintf(stderr,
"OSPRay requires the MPI runtime to support thread "
"multiple or thread serialized.\n");
return 1;
}
int mpiRank = 0;
int mpiWorldSize = 0;
MPI_Comm_rank(MPI_COMM_WORLD, &mpiRank);
MPI_Comm_size(MPI_COMM_WORLD, &mpiWorldSize);
std::cout << "OSPRay rank " << mpiRank << "/" << mpiWorldSize << "\n";
// load the MPI module, and select the MPI distributed device. Here we
// do not call ospInit, as we want to explicitly pick the distributed
// device
auto OSPRAY_MPI_DISTRIBUTED_GPU =
utility::getEnvVar<int>("OSPRAY_MPI_DISTRIBUTED_GPU").value_or(0);
if (OSPRAY_MPI_DISTRIBUTED_GPU) {
ospLoadModule("mpi_distributed_gpu");
} else {
ospLoadModule("mpi_distributed_cpu");
}
{
cpp::Device mpiDevice("mpiDistributed");
mpiDevice.commit();
mpiDevice.setCurrent();
// set an error callback to catch any OSPRay errors and exit the application
ospDeviceSetErrorCallback(
mpiDevice.handle(),
[](void *, OSPError error, const char *errorDetails) {
std::cerr << "OSPRay error: " << errorDetails << std::endl;
exit(error);
},
nullptr);
auto builder = testing::newBuilder(builderType);
testing::setParam(builder, "rendererType", rendererType);
testing::commit(builder);
auto world = testing::buildWorld(builder);
testing::release(builder);
world.commit();
cpp::Renderer renderer(rendererType);
renderer.commit();
// create a GLFW OSPRay window: this object will create and manage the
// OSPRay frame buffer and camera directly
auto glfwOSPRayWindow =
std::unique_ptr<GLFWDistribOSPRayWindow>(new GLFWDistribOSPRayWindow(
vec2i{1024, 768}, box3f(vec3f(-1.f), vec3f(1.f)), world, renderer));
// start the GLFW main loop, which will continuously render
glfwOSPRayWindow->mainLoop();
}
// cleanly shut OSPRay down
ospShutdown();
MPI_Finalize();
return 0;
}
|
