1 /*
2 * Copyright © 2010-2017 Inria. All rights reserved.
3 * Copyright © 2010-2011 Université Bordeaux
4 * Copyright © 2011 Cisco Systems, Inc. All rights reserved.
5 * See COPYING in top-level directory.
6 */
7
8 /** \file
9 * \brief Macros to help interaction between hwloc and the CUDA Driver API.
10 *
11 * Applications that use both hwloc and the CUDA Driver API may want to
12 * include this file so as to get topology information for CUDA devices.
13 *
14 */
15
16 #ifndef HWLOC_CUDA_H
17 #define HWLOC_CUDA_H
18
19 #include <hwloc.h>
20 #include <hwloc/autogen/config.h>
21 #include <hwloc/helper.h>
22 #ifdef HWLOC_LINUX_SYS
23 #include <hwloc/linux.h>
24 #endif
25
26 #include <cuda.h>
27
28
29 #ifdef __cplusplus
30 extern "C" {
31 #endif
32
33
34 /** \defgroup hwlocality_cuda Interoperability with the CUDA Driver API
35 *
36 * This interface offers ways to retrieve topology information about
37 * CUDA devices when using the CUDA Driver API.
38 *
39 * @{
40 */
41
42 /** \brief Return the domain, bus and device IDs of the CUDA device \p cudevice.
43 *
44 * Device \p cudevice must match the local machine.
45 */
46 static __hwloc_inline int
47 hwloc_cuda_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unused,
48 CUdevice cudevice, int *domain, int *bus, int *dev)
49 {
50 CUresult cres;
51
52 #if CUDA_VERSION >= 4000
53 cres = cuDeviceGetAttribute(domain, CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID, cudevice);
54 if (cres != CUDA_SUCCESS) {
55 errno = ENOSYS;
56 return -1;
57 }
58 #else
59 *domain = 0;
60 #endif
61 cres = cuDeviceGetAttribute(bus, CU_DEVICE_ATTRIBUTE_PCI_BUS_ID, cudevice);
62 if (cres != CUDA_SUCCESS) {
63 errno = ENOSYS;
64 return -1;
65 }
66 cres = cuDeviceGetAttribute(dev, CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID, cudevice);
67 if (cres != CUDA_SUCCESS) {
68 errno = ENOSYS;
69 return -1;
70 }
71
72 return 0;
73 }
74
75 /** \brief Get the CPU set of logical processors that are physically
76 * close to device \p cudevice.
77 *
78 * Return the CPU set describing the locality of the CUDA device \p cudevice.
79 *
80 * Topology \p topology and device \p cudevice must match the local machine.
81 * I/O devices detection and the CUDA component are not needed in the topology.
82 *
83 * The function only returns the locality of the device.
84 * If more information about the device is needed, OS objects should
85 * be used instead, see hwloc_cuda_get_device_osdev()
86 * and hwloc_cuda_get_device_osdev_by_index().
87 *
88 * This function is currently only implemented in a meaningful way for
89 * Linux; other systems will simply get a full cpuset.
90 */
91 static __hwloc_inline int
92 hwloc_cuda_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
93 CUdevice cudevice, hwloc_cpuset_t set)
94 {
95 #ifdef HWLOC_LINUX_SYS
96 /* If we're on Linux, use the sysfs mechanism to get the local cpus */
97 #define HWLOC_CUDA_DEVICE_SYSFS_PATH_MAX 128
98 char path[HWLOC_CUDA_DEVICE_SYSFS_PATH_MAX];
99 int domainid, busid, deviceid;
100
101 if (hwloc_cuda_get_device_pci_ids(topology, cudevice, &domainid, &busid, &deviceid))
102 return -1;
103
104 if (!hwloc_topology_is_thissystem(topology)) {
105 errno = EINVAL;
106 return -1;
107 }
108
109 sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.0/local_cpus", domainid, busid, deviceid);
110 if (hwloc_linux_read_path_as_cpumask(path, set) < 0
111 || hwloc_bitmap_iszero(set))
112 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
113 #else
114 /* Non-Linux systems simply get a full cpuset */
115 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
116 #endif
117 return 0;
118 }
119
120 /** \brief Get the hwloc PCI device object corresponding to the
121 * CUDA device \p cudevice.
122 *
123 * Return the PCI device object describing the CUDA device \p cudevice.
124 * Return NULL if there is none.
125 *
126 * Topology \p topology and device \p cudevice must match the local machine.
127 * I/O devices detection must be enabled in topology \p topology.
128 * The CUDA component is not needed in the topology.
129 */
130 static __hwloc_inline hwloc_obj_t
131 hwloc_cuda_get_device_pcidev(hwloc_topology_t topology, CUdevice cudevice)
132 {
133 int domain, bus, dev;
134
135 if (hwloc_cuda_get_device_pci_ids(topology, cudevice, &domain, &bus, &dev))
136 return NULL;
137
138 return hwloc_get_pcidev_by_busid(topology, domain, bus, dev, 0);
139 }
140
141 /** \brief Get the hwloc OS device object corresponding to CUDA device \p cudevice.
142 *
143 * Return the hwloc OS device object that describes the given
144 * CUDA device \p cudevice. Return NULL if there is none.
145 *
146 * Topology \p topology and device \p cudevice must match the local machine.
147 * I/O devices detection and the CUDA component must be enabled in the topology.
148 * If not, the locality of the object may still be found using
149 * hwloc_cuda_get_device_cpuset().
150 *
151 * \note This function cannot work if PCI devices are filtered out.
152 *
153 * \note The corresponding hwloc PCI device may be found by looking
154 * at the result parent pointer (unless PCI devices are filtered out).
155 */
156 static __hwloc_inline hwloc_obj_t
157 hwloc_cuda_get_device_osdev(hwloc_topology_t topology, CUdevice cudevice)
158 {
159 hwloc_obj_t osdev = NULL;
160 int domain, bus, dev;
161
162 if (hwloc_cuda_get_device_pci_ids(topology, cudevice, &domain, &bus, &dev))
163 return NULL;
164
165 osdev = NULL;
166 while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
167 hwloc_obj_t pcidev = osdev->parent;
168 if (strncmp(osdev->name, "cuda", 4))
169 continue;
170 if (pcidev
171 && pcidev->type == HWLOC_OBJ_PCI_DEVICE
172 && (int) pcidev->attr->pcidev.domain == domain
173 && (int) pcidev->attr->pcidev.bus == bus
174 && (int) pcidev->attr->pcidev.dev == dev
175 && pcidev->attr->pcidev.func == 0)
176 return osdev;
177 /* if PCI are filtered out, we need a info attr to match on */
178 }
179
180 return NULL;
181 }
182
183 /** \brief Get the hwloc OS device object corresponding to the
184 * CUDA device whose index is \p idx.
185 *
186 * Return the OS device object describing the CUDA device whose
187 * index is \p idx. Return NULL if there is none.
188 *
189 * The topology \p topology does not necessarily have to match the current
190 * machine. For instance the topology may be an XML import of a remote host.
191 * I/O devices detection and the CUDA component must be enabled in the topology.
192 *
193 * \note The corresponding PCI device object can be obtained by looking
194 * at the OS device parent object (unless PCI devices are filtered out).
195 *
196 * \note This function is identical to hwloc_cudart_get_device_osdev_by_index().
197 */
198 static __hwloc_inline hwloc_obj_t
199 hwloc_cuda_get_device_osdev_by_index(hwloc_topology_t topology, unsigned idx)
200 {
201 hwloc_obj_t osdev = NULL;
202 while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
203 if (HWLOC_OBJ_OSDEV_COPROC == osdev->attr->osdev.type
204 && osdev->name
205 && !strncmp("cuda", osdev->name, 4)
206 && atoi(osdev->name + 4) == (int) idx)
207 return osdev;
208 }
209 return NULL;
210 }
211
212 /** @} */
213
214
215 #ifdef __cplusplus
216 } /* extern "C" */
217 #endif
218
219
220 #endif /* HWLOC_CUDA_H */