1 /*
2 * Copyright © 2012-2021 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 #else
27 #include <CL/cl.h>
28 #endif
29
30 #include <stdio.h>
31
32
33 #ifdef __cplusplus
34 extern "C" {
35 #endif
36
37
38 /* OpenCL extensions aren't always shipped with default headers, and
39 * they don't always reflect what the installed implementations support.
40 * Try everything and let the implementation return errors when non supported.
41 */
42 /* Copyright (c) 2008-2018 The Khronos Group Inc. */
43
44 /* needs "cl_amd_device_attribute_query" device extension, but not strictly required for clGetDeviceInfo() */
45 #define HWLOC_CL_DEVICE_TOPOLOGY_AMD 0x4037
46 typedef union {
47 struct { cl_uint type; cl_uint data[5]; } raw;
48 struct { cl_uint type; cl_char unused[17]; cl_char bus; cl_char device; cl_char function; } pcie;
49 } hwloc_cl_device_topology_amd;
50 #define HWLOC_CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD 1
51
52 /* needs "cl_nv_device_attribute_query" device extension, but not strictly required for clGetDeviceInfo() */
53 #define HWLOC_CL_DEVICE_PCI_BUS_ID_NV 0x4008
54 #define HWLOC_CL_DEVICE_PCI_SLOT_ID_NV 0x4009
55 #define HWLOC_CL_DEVICE_PCI_DOMAIN_ID_NV 0x400A
56
57
58 /** \defgroup hwlocality_opencl Interoperability with OpenCL
59 *
60 * This interface offers ways to retrieve topology information about
61 * OpenCL devices.
62 *
63 * Only AMD and NVIDIA OpenCL implementations currently offer useful locality
64 * information about their devices.
65 *
66 * @{
67 */
68
69 /** \brief Return the domain, bus and device IDs of the OpenCL device \p device.
70 *
71 * Device \p device must match the local machine.
72 */
73 static __hwloc_inline int
hwloc_opencl_get_device_pci_busid(cl_device_id device,unsigned * domain,unsigned * bus,unsigned * dev,unsigned * func)74 hwloc_opencl_get_device_pci_busid(cl_device_id device,
75 unsigned *domain, unsigned *bus, unsigned *dev, unsigned *func)
76 {
77 hwloc_cl_device_topology_amd amdtopo;
78 cl_uint nvbus, nvslot, nvdomain;
79 cl_int clret;
80
81 clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_TOPOLOGY_AMD, sizeof(amdtopo), &amdtopo, NULL);
82 if (CL_SUCCESS == clret
83 && HWLOC_CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD == amdtopo.raw.type) {
84 *domain = 0; /* can't do anything better */
85 /* cl_device_topology_amd stores bus ID in cl_char, dont convert those signed char directly to unsigned int */
86 *bus = (unsigned) (unsigned char) amdtopo.pcie.bus;
87 *dev = (unsigned) (unsigned char) amdtopo.pcie.device;
88 *func = (unsigned) (unsigned char) amdtopo.pcie.function;
89 return 0;
90 }
91
92 clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_BUS_ID_NV, sizeof(nvbus), &nvbus, NULL);
93 if (CL_SUCCESS == clret) {
94 clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_SLOT_ID_NV, sizeof(nvslot), &nvslot, NULL);
95 if (CL_SUCCESS == clret) {
96 clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_DOMAIN_ID_NV, sizeof(nvdomain), &nvdomain, NULL);
97 if (CL_SUCCESS == clret) { /* available since CUDA 10.2 */
98 *domain = nvdomain;
99 } else {
100 *domain = 0;
101 }
102 *bus = nvbus & 0xff;
103 /* non-documented but used in many other projects */
104 *dev = nvslot >> 3;
105 *func = nvslot & 0x7;
106 return 0;
107 }
108 }
109
110 return -1;
111 }
112
113 /** \brief Get the CPU set of processors that are physically
114 * close to OpenCL device \p device.
115 *
116 * Store in \p set the CPU-set describing the locality of the OpenCL device \p device.
117 *
118 * Topology \p topology and device \p device must match the local machine.
119 * I/O devices detection and the OpenCL component are not needed in the topology.
120 *
121 * The function only returns the locality of the device.
122 * If more information about the device is needed, OS objects should
123 * be used instead, see hwloc_opencl_get_device_osdev()
124 * and hwloc_opencl_get_device_osdev_by_index().
125 *
126 * This function is currently only implemented in a meaningful way for
127 * Linux with the AMD or NVIDIA OpenCL implementation; other systems will simply
128 * get a full cpuset.
129 */
130 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)131 hwloc_opencl_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
132 cl_device_id device __hwloc_attribute_unused,
133 hwloc_cpuset_t set)
134 {
135 #if (defined HWLOC_LINUX_SYS)
136 /* If we're on Linux, try AMD/NVIDIA extensions + the sysfs mechanism to get the local cpus */
137 #define HWLOC_OPENCL_DEVICE_SYSFS_PATH_MAX 128
138 char path[HWLOC_OPENCL_DEVICE_SYSFS_PATH_MAX];
139 unsigned pcidomain, pcibus, pcidev, pcifunc;
140
141 if (!hwloc_topology_is_thissystem(topology)) {
142 errno = EINVAL;
143 return -1;
144 }
145
146 if (hwloc_opencl_get_device_pci_busid(device, &pcidomain, &pcibus, &pcidev, &pcifunc) < 0) {
147 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
148 return 0;
149 }
150
151 sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.%01x/local_cpus", pcidomain, pcibus, pcidev, pcifunc);
152 if (hwloc_linux_read_path_as_cpumask(path, set) < 0
153 || hwloc_bitmap_iszero(set))
154 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
155 #else
156 /* Non-Linux systems simply get a full cpuset */
157 hwloc_bitmap_copy(set, hwloc_topology_get_complete_cpuset(topology));
158 #endif
159 return 0;
160 }
161
162 /** \brief Get the hwloc OS device object corresponding to the
163 * OpenCL device for the given indexes.
164 *
165 * \return The hwloc OS device object describing the OpenCL device
166 * whose platform index is \p platform_index,
167 * and whose device index within this platform if \p device_index.
168 * \return \c NULL if there is none.
169 *
170 * The topology \p topology does not necessarily have to match the current
171 * machine. For instance the topology may be an XML import of a remote host.
172 * I/O devices detection and the OpenCL component must be enabled in the topology.
173 *
174 * \note The corresponding PCI device object can be obtained by looking
175 * at the OS device parent object (unless PCI devices are filtered out).
176 */
177 static __hwloc_inline hwloc_obj_t
hwloc_opencl_get_device_osdev_by_index(hwloc_topology_t topology,unsigned platform_index,unsigned device_index)178 hwloc_opencl_get_device_osdev_by_index(hwloc_topology_t topology,
179 unsigned platform_index, unsigned device_index)
180 {
181 unsigned x = (unsigned) -1, y = (unsigned) -1;
182 hwloc_obj_t osdev = NULL;
183 while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
184 if (HWLOC_OBJ_OSDEV_COPROC == osdev->attr->osdev.type
185 && osdev->name
186 && sscanf(osdev->name, "opencl%ud%u", &x, &y) == 2
187 && platform_index == x && device_index == y)
188 return osdev;
189 }
190 return NULL;
191 }
192
193 /** \brief Get the hwloc OS device object corresponding to OpenCL device \p deviceX.
194 *
195 * \return The hwloc OS device object corresponding to the given OpenCL device \p device.
196 * \return \c NULL if none could be found, for instance
197 * if required OpenCL attributes are not available.
198 *
199 * This function currently only works on AMD and NVIDIA OpenCL devices that support
200 * relevant OpenCL extensions. hwloc_opencl_get_device_osdev_by_index()
201 * should be preferred whenever possible, i.e. when platform and device index
202 * are known.
203 *
204 * Topology \p topology and device \p device must match the local machine.
205 * I/O devices detection and the OpenCL component must be enabled in the topology.
206 * If not, the locality of the object may still be found using
207 * hwloc_opencl_get_device_cpuset().
208 *
209 * \note This function cannot work if PCI devices are filtered out.
210 *
211 * \note The corresponding hwloc PCI device may be found by looking
212 * at the result parent pointer (unless PCI devices are filtered out).
213 */
214 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)215 hwloc_opencl_get_device_osdev(hwloc_topology_t topology __hwloc_attribute_unused,
216 cl_device_id device __hwloc_attribute_unused)
217 {
218 hwloc_obj_t osdev;
219 unsigned pcidomain, pcibus, pcidevice, pcifunc;
220
221 if (hwloc_opencl_get_device_pci_busid(device, &pcidomain, &pcibus, &pcidevice, &pcifunc) < 0) {
222 errno = EINVAL;
223 return NULL;
224 }
225
226 osdev = NULL;
227 while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
228 hwloc_obj_t pcidev = osdev->parent;
229 if (strncmp(osdev->name, "opencl", 6))
230 continue;
231 if (pcidev
232 && pcidev->type == HWLOC_OBJ_PCI_DEVICE
233 && pcidev->attr->pcidev.domain == pcidomain
234 && pcidev->attr->pcidev.bus == pcibus
235 && pcidev->attr->pcidev.dev == pcidevice
236 && pcidev->attr->pcidev.func == pcifunc)
237 return osdev;
238 /* if PCI are filtered out, we need a info attr to match on */
239 }
240
241 return NULL;
242 }
243
244 /** @} */
245
246
247 #ifdef __cplusplus
248 } /* extern "C" */
249 #endif
250
251
252 #endif /* HWLOC_OPENCL_H */
253