1 /*
2 * Copyright © 2012-2017 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
hwloc_opencl_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,cl_device_id device __hwloc_attribute_unused,hwloc_cpuset_t set)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 FILE *sysfile = NULL;
78 cl_device_topology_amd amdtopo;
79 cl_int clret;
80
81 if (!hwloc_topology_is_thissystem(topology)) {
82 errno = EINVAL;
83 return -1;
84 }
85
86 clret = clGetDeviceInfo(device, CL_DEVICE_TOPOLOGY_AMD, sizeof(amdtopo), &amdtopo, NULL);
87 if (CL_SUCCESS != clret) {
88 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
89 return 0;
90 }
91 if (CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD != amdtopo.raw.type) {
92 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
93 return 0;
94 }
95
96 sprintf(path, "/sys/bus/pci/devices/0000:%02x:%02x.%01x/local_cpus",
97 (unsigned) amdtopo.pcie.bus, (unsigned) amdtopo.pcie.device, (unsigned) amdtopo.pcie.function);
98 sysfile = fopen(path, "r");
99 if (!sysfile)
100 return -1;
101
102 if (hwloc_linux_parse_cpumap_file(sysfile, set) < 0
103 || hwloc_bitmap_iszero(set))
104 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
105
106 fclose(sysfile);
107 #else
108 /* Non-Linux + AMD OpenCL systems simply get a full cpuset */
109 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
110 #endif
111 return 0;
112 }
113
114 /** \brief Get the hwloc OS device object corresponding to the
115 * OpenCL device for the given indexes.
116 *
117 * Return the OS device object describing the OpenCL device
118 * whose platform index is \p platform_index,
119 * and whose device index within this platform if \p device_index.
120 * Return NULL if there is none.
121 *
122 * The topology \p topology does not necessarily have to match the current
123 * machine. For instance the topology may be an XML import of a remote host.
124 * I/O devices detection and the OpenCL component must be enabled in the topology.
125 *
126 * \note The corresponding PCI device object can be obtained by looking
127 * at the OS device parent object.
128 */
129 static __hwloc_inline hwloc_obj_t
hwloc_opencl_get_device_osdev_by_index(hwloc_topology_t topology,unsigned platform_index,unsigned device_index)130 hwloc_opencl_get_device_osdev_by_index(hwloc_topology_t topology,
131 unsigned platform_index, unsigned device_index)
132 {
133 unsigned x = (unsigned) -1, y = (unsigned) -1;
134 hwloc_obj_t osdev = NULL;
135 while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
136 if (HWLOC_OBJ_OSDEV_COPROC == osdev->attr->osdev.type
137 && osdev->name
138 && sscanf(osdev->name, "opencl%ud%u", &x, &y) == 2
139 && platform_index == x && device_index == y)
140 return osdev;
141 }
142 return NULL;
143 }
144
145 /** \brief Get the hwloc OS device object corresponding to OpenCL device \p device.
146 *
147 * Return the hwloc OS device object that describes the given
148 * OpenCL device \p device. Return NULL if there is none.
149 *
150 * Topology \p topology and device \p device must match the local machine.
151 * I/O devices detection and the OpenCL component must be enabled in the topology.
152 * If not, the locality of the object may still be found using
153 * hwloc_opencl_get_device_cpuset().
154 *
155 * \note The corresponding hwloc PCI device may be found by looking
156 * at the result parent pointer.
157 */
158 static __hwloc_inline hwloc_obj_t
hwloc_opencl_get_device_osdev(hwloc_topology_t topology __hwloc_attribute_unused,cl_device_id device __hwloc_attribute_unused)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 }
190
191 return NULL;
192 #else
193 return NULL;
194 #endif
195 }
196
197 /** @} */
198
199
200 #ifdef __cplusplus
201 } /* extern "C" */
202 #endif
203
204
205 #endif /* HWLOC_OPENCL_H */
206