1 /*
2 * Copyright © 2012-2018 Inria. All rights reserved.
3 * Copyright © 2013, 2018 Université Bordeaux. All right reserved.
4 * See COPYING in top-level directory.
5 */
6
7 /** \file
8 * \brief Macros to help interaction between hwloc and the OpenCL interface.
9 *
10 * Applications that use both hwloc and OpenCL may want to
11 * include this file so as to get topology information for OpenCL devices.
12 */
13
14 #ifndef HWLOC_OPENCL_H
15 #define HWLOC_OPENCL_H
16
17 #include <hwloc.h>
18 #include <hwloc/autogen/config.h>
19 #include <hwloc/helper.h>
20 #ifdef HWLOC_LINUX_SYS
21 #include <hwloc/linux.h>
22 #endif
23
24 #ifdef __APPLE__
25 #include <OpenCL/cl.h>
26 #include <OpenCL/cl_ext.h>
27 #else
28 #include <CL/cl.h>
29 #include <CL/cl_ext.h>
30 #endif
31
32 #include <stdio.h>
33
34
35 #ifdef __cplusplus
36 extern "C" {
37 #endif
38
39
40 /** \defgroup hwlocality_opencl Interoperability with OpenCL
41 *
42 * This interface offers ways to retrieve topology information about
43 * OpenCL devices.
44 *
45 * Only the AMD OpenCL interface currently offers useful locality information
46 * about its devices.
47 *
48 * @{
49 */
50
51 /** \brief Get the CPU set of logical processors that are physically
52 * close to OpenCL device \p device.
53 *
54 * Return the CPU set describing the locality of the OpenCL device \p device.
55 *
56 * Topology \p topology and device \p device must match the local machine.
57 * I/O devices detection and the OpenCL component are not needed in the topology.
58 *
59 * The function only returns the locality of the device.
60 * If more information about the device is needed, OS objects should
61 * be used instead, see hwloc_opencl_get_device_osdev()
62 * and hwloc_opencl_get_device_osdev_by_index().
63 *
64 * This function is currently only implemented in a meaningful way for
65 * Linux with the AMD OpenCL implementation; other systems will simply
66 * get a full cpuset.
67 */
68 static __hwloc_inline int
69 hwloc_opencl_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
70 cl_device_id device __hwloc_attribute_unused,
71 hwloc_cpuset_t set)
72 {
73 #if (defined HWLOC_LINUX_SYS) && (defined CL_DEVICE_TOPOLOGY_AMD)
74 /* If we're on Linux + AMD OpenCL, use the AMD extension + the sysfs mechanism to get the local cpus */
75 #define HWLOC_OPENCL_DEVICE_SYSFS_PATH_MAX 128
76 char path[HWLOC_OPENCL_DEVICE_SYSFS_PATH_MAX];
77 cl_device_topology_amd amdtopo;
78 cl_int clret;
79
80 if (!hwloc_topology_is_thissystem(topology)) {
81 errno = EINVAL;
82 return -1;
83 }
84
85 clret = clGetDeviceInfo(device, CL_DEVICE_TOPOLOGY_AMD, sizeof(amdtopo), &amdtopo, NULL);
86 if (CL_SUCCESS != clret) {
87 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
88 return 0;
89 }
90 if (CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD != amdtopo.raw.type) {
91 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
92 return 0;
93 }
94
95 sprintf(path, "/sys/bus/pci/devices/0000:%02x:%02x.%01x/local_cpus",
96 (unsigned) amdtopo.pcie.bus, (unsigned) amdtopo.pcie.device, (unsigned) amdtopo.pcie.function);
97 if (hwloc_linux_read_path_as_cpumask(path, set) < 0
98 || hwloc_bitmap_iszero(set))
99 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
100 #else
101 /* Non-Linux + AMD OpenCL systems simply get a full cpuset */
102 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
103 #endif
104 return 0;
105 }
106
107 /** \brief Get the hwloc OS device object corresponding to the
108 * OpenCL device for the given indexes.
109 *
110 * Return the OS device object describing the OpenCL device
111 * whose platform index is \p platform_index,
112 * and whose device index within this platform if \p device_index.
113 * Return NULL if there is none.
114 *
115 * The topology \p topology does not necessarily have to match the current
116 * machine. For instance the topology may be an XML import of a remote host.
117 * I/O devices detection and the OpenCL component must be enabled in the topology.
118 *
119 * \note The corresponding PCI device object can be obtained by looking
120 * at the OS device parent object (unless PCI devices are filtered out).
121 */
122 static __hwloc_inline hwloc_obj_t
123 hwloc_opencl_get_device_osdev_by_index(hwloc_topology_t topology,
124 unsigned platform_index, unsigned device_index)
125 {
126 unsigned x = (unsigned) -1, y = (unsigned) -1;
127 hwloc_obj_t osdev = NULL;
128 while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
129 if (HWLOC_OBJ_OSDEV_COPROC == osdev->attr->osdev.type
130 && osdev->name
131 && sscanf(osdev->name, "opencl%ud%u", &x, &y) == 2
132 && platform_index == x && device_index == y)
133 return osdev;
134 }
135 return NULL;
136 }
137
138 /** \brief Get the hwloc OS device object corresponding to OpenCL device \p deviceX.
139 *
140 * Use OpenCL device attributes to find the corresponding hwloc OS device object.
141 * Return NULL if there is none or if useful attributes are not available.
142 *
143 * This function currently only works on AMD OpenCL devices that support
144 * the CL_DEVICE_TOPOLOGY_AMD extension. hwloc_opencl_get_device_osdev_by_index()
145 * should be preferred whenever possible, i.e. when platform and device index
146 * are known.
147 *
148 * Topology \p topology and device \p device must match the local machine.
149 * I/O devices detection and the OpenCL component must be enabled in the topology.
150 * If not, the locality of the object may still be found using
151 * hwloc_opencl_get_device_cpuset().
152 *
153 * \note This function cannot work if PCI devices are filtered out.
154 *
155 * \note The corresponding hwloc PCI device may be found by looking
156 * at the result parent pointer (unless PCI devices are filtered out).
157 */
158 static __hwloc_inline hwloc_obj_t
159 hwloc_opencl_get_device_osdev(hwloc_topology_t topology __hwloc_attribute_unused,
160 cl_device_id device __hwloc_attribute_unused)
161 {
162 #ifdef CL_DEVICE_TOPOLOGY_AMD
163 hwloc_obj_t osdev;
164 cl_device_topology_amd amdtopo;
165 cl_int clret;
166
167 clret = clGetDeviceInfo(device, CL_DEVICE_TOPOLOGY_AMD, sizeof(amdtopo), &amdtopo, NULL);
168 if (CL_SUCCESS != clret) {
169 errno = EINVAL;
170 return NULL;
171 }
172 if (CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD != amdtopo.raw.type) {
173 errno = EINVAL;
174 return NULL;
175 }
176
177 osdev = NULL;
178 while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
179 hwloc_obj_t pcidev = osdev->parent;
180 if (strncmp(osdev->name, "opencl", 6))
181 continue;
182 if (pcidev
183 && pcidev->type == HWLOC_OBJ_PCI_DEVICE
184 && pcidev->attr->pcidev.domain == 0
185 && pcidev->attr->pcidev.bus == amdtopo.pcie.bus
186 && pcidev->attr->pcidev.dev == amdtopo.pcie.device
187 && pcidev->attr->pcidev.func == amdtopo.pcie.function)
188 return osdev;
189 /* if PCI are filtered out, we need a info attr to match on */
190 }
191
192 return NULL;
193 #else
194 return NULL;
195 #endif
196 }
197
198 /** @} */
199
200
201 #ifdef __cplusplus
202 } /* extern "C" */
203 #endif
204
205
206 #endif /* HWLOC_OPENCL_H */