Hardware Locality (hwloc) 2.9.0
opencl.h
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
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
34extern "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
46typedef 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;
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
73static __hwloc_inline int
75 unsigned *domain, unsigned *bus, unsigned *dev, unsigned *func)
76{
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
130static __hwloc_inline int
131hwloc_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) {
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))
155#else
156 /* Non-Linux systems simply get a full cpuset */
158#endif
159 return 0;
160}
161
177static __hwloc_inline hwloc_obj_t
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) {
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
214static __hwloc_inline hwloc_obj_t
215hwloc_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
247#ifdef __cplusplus
248} /* extern "C" */
249#endif
250
251
252#endif /* HWLOC_OPENCL_H */
hwloc_bitmap_t hwloc_cpuset_t
A CPU set is a bitmap whose bits are set according to CPU physical OS indexes.
Definition: hwloc.h:140
@ HWLOC_OBJ_OSDEV_COPROC
Operating system co-processor device. For instance "opencl0d0" for a OpenCL device,...
Definition: hwloc.h:353
@ HWLOC_OBJ_PCI_DEVICE
PCI device (filtered out by default).
Definition: hwloc.h:269
struct hwloc_topology * hwloc_topology_t
Topology context.
Definition: hwloc.h:692
int hwloc_topology_is_thissystem(hwloc_topology_t restrict topology)
Does the topology context come from this system?
hwloc_const_cpuset_t hwloc_topology_get_complete_cpuset(hwloc_topology_t topology)
Get complete CPU set.
static hwloc_obj_t hwloc_get_next_osdev(hwloc_topology_t topology, hwloc_obj_t prev)
Get the next OS device in the system.
Definition: helper.h:1191
int hwloc_bitmap_iszero(hwloc_const_bitmap_t bitmap)
Test whether bitmap bitmap is empty.
int hwloc_bitmap_copy(hwloc_bitmap_t dst, hwloc_const_bitmap_t src)
Copy the contents of bitmap src into the already allocated bitmap dst.
int hwloc_linux_read_path_as_cpumask(const char *path, hwloc_bitmap_t set)
Convert a linux kernel cpumask file path into a hwloc bitmap set.
static int hwloc_opencl_get_device_pci_busid(cl_device_id device, unsigned *domain, unsigned *bus, unsigned *dev, unsigned *func)
Return the domain, bus and device IDs of the OpenCL device device.
Definition: opencl.h:74
static int hwloc_opencl_get_device_cpuset(hwloc_topology_t topology, cl_device_id device, hwloc_cpuset_t set)
Get the CPU set of processors that are physically close to OpenCL device device.
Definition: opencl.h:131
static hwloc_obj_t hwloc_opencl_get_device_osdev(hwloc_topology_t topology, cl_device_id device)
Get the hwloc OS device object corresponding to OpenCL device deviceX.
Definition: opencl.h:215
static hwloc_obj_t hwloc_opencl_get_device_osdev_by_index(hwloc_topology_t topology, unsigned platform_index, unsigned device_index)
Get the hwloc OS device object corresponding to the OpenCL device for the given indexes.
Definition: opencl.h:178
Structure of a topology object.
Definition: hwloc.h:396
char * name
Object-specific name if any. Mostly used for identifying OS devices and Misc objects where a name str...
Definition: hwloc.h:408
hwloc_obj_type_t type
Type of object.
Definition: hwloc.h:398
union hwloc_obj_attr_u * attr
Object type-specific Attributes, may be NULL if no attribute value was found.
Definition: hwloc.h:415
struct hwloc_obj * parent
Parent, NULL if root (Machine object)
Definition: hwloc.h:446
struct hwloc_obj_attr_u::hwloc_pcidev_attr_s pcidev
struct hwloc_obj_attr_u::hwloc_osdev_attr_s osdev
unsigned char dev
Definition: hwloc.h:639
unsigned char func
Definition: hwloc.h:639
unsigned short domain
Definition: hwloc.h:635
unsigned char bus
Definition: hwloc.h:639
hwloc_obj_osdev_type_t type
Definition: hwloc.h:666
Definition: opencl.h:46
cl_char device
Definition: opencl.h:48
struct hwloc_cl_device_topology_amd::@4 pcie
cl_char bus
Definition: opencl.h:48
cl_char function
Definition: opencl.h:48
struct hwloc_cl_device_topology_amd::@3 raw
cl_uint type
Definition: opencl.h:47