Hardware Locality (hwloc)  2.10.0
opencl.h
1 /*
2  * Copyright © 2012-2023 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
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;
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 
76 static __hwloc_inline int
78  unsigned *domain, unsigned *bus, unsigned *dev, unsigned *func)
79 {
81  cl_uint nvbus, nvslot, nvdomain;
82  cl_int clret;
83 
84  clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_TOPOLOGY_AMD, sizeof(amdtopo), &amdtopo, NULL);
85  if (CL_SUCCESS == clret
86  && HWLOC_CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD == amdtopo.raw.type) {
87  *domain = 0; /* can't do anything better */
88  /* cl_device_topology_amd stores bus ID in cl_char, dont convert those signed char directly to unsigned int */
89  *bus = (unsigned) (unsigned char) amdtopo.pcie.bus;
90  *dev = (unsigned) (unsigned char) amdtopo.pcie.device;
91  *func = (unsigned) (unsigned char) amdtopo.pcie.function;
92  return 0;
93  }
94 
95  clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_BUS_ID_NV, sizeof(nvbus), &nvbus, NULL);
96  if (CL_SUCCESS == clret) {
97  clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_SLOT_ID_NV, sizeof(nvslot), &nvslot, NULL);
98  if (CL_SUCCESS == clret) {
99  clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_DOMAIN_ID_NV, sizeof(nvdomain), &nvdomain, NULL);
100  if (CL_SUCCESS == clret) { /* available since CUDA 10.2 */
101  *domain = nvdomain;
102  } else {
103  *domain = 0;
104  }
105  *bus = nvbus & 0xff;
106  /* non-documented but used in many other projects */
107  *dev = nvslot >> 3;
108  *func = nvslot & 0x7;
109  return 0;
110  }
111  }
112 
113  return -1;
114 }
115 
136 static __hwloc_inline int
137 hwloc_opencl_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
138  cl_device_id device __hwloc_attribute_unused,
139  hwloc_cpuset_t set)
140 {
141 #if (defined HWLOC_LINUX_SYS)
142  /* If we're on Linux, try AMD/NVIDIA extensions + the sysfs mechanism to get the local cpus */
143 #define HWLOC_OPENCL_DEVICE_SYSFS_PATH_MAX 128
144  char path[HWLOC_OPENCL_DEVICE_SYSFS_PATH_MAX];
145  unsigned pcidomain, pcibus, pcidev, pcifunc;
146 
147  if (!hwloc_topology_is_thissystem(topology)) {
148  errno = EINVAL;
149  return -1;
150  }
151 
152  if (hwloc_opencl_get_device_pci_busid(device, &pcidomain, &pcibus, &pcidev, &pcifunc) < 0) {
154  return 0;
155  }
156 
157  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.%01x/local_cpus", pcidomain, pcibus, pcidev, pcifunc);
158  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
159  || hwloc_bitmap_iszero(set))
161 #else
162  /* Non-Linux systems simply get a full cpuset */
164 #endif
165  return 0;
166 }
167 
183 static __hwloc_inline hwloc_obj_t
185  unsigned platform_index, unsigned device_index)
186 {
187  unsigned x = (unsigned) -1, y = (unsigned) -1;
188  hwloc_obj_t osdev = NULL;
189  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
190  if (HWLOC_OBJ_OSDEV_COPROC == osdev->attr->osdev.type
191  && osdev->name
192  && sscanf(osdev->name, "opencl%ud%u", &x, &y) == 2
193  && platform_index == x && device_index == y)
194  return osdev;
195  }
196  return NULL;
197 }
198 
220 static __hwloc_inline hwloc_obj_t
221 hwloc_opencl_get_device_osdev(hwloc_topology_t topology __hwloc_attribute_unused,
222  cl_device_id device __hwloc_attribute_unused)
223 {
224  hwloc_obj_t osdev;
225  unsigned pcidomain, pcibus, pcidevice, pcifunc;
226 
227  if (hwloc_opencl_get_device_pci_busid(device, &pcidomain, &pcibus, &pcidevice, &pcifunc) < 0) {
228  errno = EINVAL;
229  return NULL;
230  }
231 
232  osdev = NULL;
233  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
234  hwloc_obj_t pcidev = osdev->parent;
235  if (strncmp(osdev->name, "opencl", 6))
236  continue;
237  if (pcidev
238  && pcidev->type == HWLOC_OBJ_PCI_DEVICE
239  && pcidev->attr->pcidev.domain == pcidomain
240  && pcidev->attr->pcidev.bus == pcibus
241  && pcidev->attr->pcidev.dev == pcidevice
242  && pcidev->attr->pcidev.func == pcifunc)
243  return osdev;
244  /* if PCI are filtered out, we need a info attr to match on */
245  }
246 
247  return NULL;
248 }
249 
253 #ifdef __cplusplus
254 } /* extern "C" */
255 #endif
256 
257 
258 #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:161
@ HWLOC_OBJ_OSDEV_COPROC
Operating system co-processor device. For instance "opencl0d0" for a OpenCL device,...
Definition: hwloc.h:385
@ HWLOC_OBJ_PCI_DEVICE
PCI device (filtered out by default).
Definition: hwloc.h:295
struct hwloc_topology * hwloc_topology_t
Topology context.
Definition: hwloc.h:742
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:1282
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:77
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:137
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:221
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:184
Structure of a topology object.
Definition: hwloc.h:431
char * name
Object-specific name if any. Mostly used for identifying OS devices and Misc objects where a name str...
Definition: hwloc.h:443
hwloc_obj_type_t type
Type of object.
Definition: hwloc.h:433
union hwloc_obj_attr_u * attr
Object type-specific Attributes, may be NULL if no attribute value was found.
Definition: hwloc.h:450
struct hwloc_obj * parent
Parent, NULL if root (Machine object)
Definition: hwloc.h:481
struct hwloc_obj_attr_u::hwloc_pcidev_attr_s pcidev
struct hwloc_obj_attr_u::hwloc_osdev_attr_s osdev
unsigned char dev
Device number (zz in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:677
unsigned char func
Function number (t in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:678
unsigned short domain
Domain number (xxxx in the PCI BDF notation xxxx:yy:zz.t). Only 16bits PCI domains are supported by d...
Definition: hwloc.h:670
unsigned char bus
Bus number (yy in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:676
hwloc_obj_osdev_type_t type
Definition: hwloc.h:716
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