Hardware Locality (hwloc)  2.10.0
cuda.h
1 /*
2  * Copyright © 2010-2023 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 
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 
49 static __hwloc_inline int
50 hwloc_cuda_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unused,
51  CUdevice cudevice, int *domain, int *bus, int *dev)
52 {
53  CUresult cres;
54 
55 #if CUDA_VERSION >= 4000
56  cres = cuDeviceGetAttribute(domain, CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID, cudevice);
57  if (cres != CUDA_SUCCESS) {
58  errno = ENOSYS;
59  return -1;
60  }
61 #else
62  *domain = 0;
63 #endif
64  cres = cuDeviceGetAttribute(bus, CU_DEVICE_ATTRIBUTE_PCI_BUS_ID, cudevice);
65  if (cres != CUDA_SUCCESS) {
66  errno = ENOSYS;
67  return -1;
68  }
69  cres = cuDeviceGetAttribute(dev, CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID, cudevice);
70  if (cres != CUDA_SUCCESS) {
71  errno = ENOSYS;
72  return -1;
73  }
74 
75  return 0;
76 }
77 
97 static __hwloc_inline int
98 hwloc_cuda_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
99  CUdevice cudevice, hwloc_cpuset_t set)
100 {
101 #ifdef HWLOC_LINUX_SYS
102  /* If we're on Linux, use the sysfs mechanism to get the local cpus */
103 #define HWLOC_CUDA_DEVICE_SYSFS_PATH_MAX 128
104  char path[HWLOC_CUDA_DEVICE_SYSFS_PATH_MAX];
105  int domainid, busid, deviceid;
106 
107  if (hwloc_cuda_get_device_pci_ids(topology, cudevice, &domainid, &busid, &deviceid))
108  return -1;
109 
110  if (!hwloc_topology_is_thissystem(topology)) {
111  errno = EINVAL;
112  return -1;
113  }
114 
115  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.0/local_cpus", domainid, busid, deviceid);
116  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
117  || hwloc_bitmap_iszero(set))
119 #else
120  /* Non-Linux systems simply get a full cpuset */
122 #endif
123  return 0;
124 }
125 
136 static __hwloc_inline hwloc_obj_t
137 hwloc_cuda_get_device_pcidev(hwloc_topology_t topology, CUdevice cudevice)
138 {
139  int domain, bus, dev;
140 
141  if (hwloc_cuda_get_device_pci_ids(topology, cudevice, &domain, &bus, &dev))
142  return NULL;
143 
144  return hwloc_get_pcidev_by_busid(topology, domain, bus, dev, 0);
145 }
146 
162 static __hwloc_inline hwloc_obj_t
163 hwloc_cuda_get_device_osdev(hwloc_topology_t topology, CUdevice cudevice)
164 {
165  hwloc_obj_t osdev = NULL;
166  int domain, bus, dev;
167 
168  if (hwloc_cuda_get_device_pci_ids(topology, cudevice, &domain, &bus, &dev))
169  return NULL;
170 
171  osdev = NULL;
172  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
173  hwloc_obj_t pcidev = osdev->parent;
174  if (strncmp(osdev->name, "cuda", 4))
175  continue;
176  if (pcidev
177  && pcidev->type == HWLOC_OBJ_PCI_DEVICE
178  && (int) pcidev->attr->pcidev.domain == domain
179  && (int) pcidev->attr->pcidev.bus == bus
180  && (int) pcidev->attr->pcidev.dev == dev
181  && pcidev->attr->pcidev.func == 0)
182  return osdev;
183  /* if PCI are filtered out, we need a info attr to match on */
184  }
185 
186  return NULL;
187 }
188 
204 static __hwloc_inline hwloc_obj_t
206 {
207  hwloc_obj_t osdev = NULL;
208  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
209  if (HWLOC_OBJ_OSDEV_COPROC == osdev->attr->osdev.type
210  && osdev->name
211  && !strncmp("cuda", osdev->name, 4)
212  && atoi(osdev->name + 4) == (int) idx)
213  return osdev;
214  }
215  return NULL;
216 }
217 
221 #ifdef __cplusplus
222 } /* extern "C" */
223 #endif
224 
225 
226 #endif /* HWLOC_CUDA_H */
unsigned char func
Function number (t in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:678
static int hwloc_cuda_get_device_cpuset(hwloc_topology_t topology, CUdevice cudevice, hwloc_cpuset_t set)
Get the CPU set of processors that are physically close to device cudevice.
Definition: cuda.h:98
static int hwloc_cuda_get_device_pci_ids(hwloc_topology_t topology, CUdevice cudevice, int *domain, int *bus, int *dev)
Return the domain, bus and device IDs of the CUDA device cudevice.
Definition: cuda.h:50
Structure of a topology object.
Definition: hwloc.h:431
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_copy(hwloc_bitmap_t dst, hwloc_const_bitmap_t src)
Copy the contents of bitmap src into the already allocated bitmap dst.
struct hwloc_obj_attr_u::hwloc_pcidev_attr_s pcidev
hwloc_const_cpuset_t hwloc_topology_get_complete_cpuset(hwloc_topology_t topology)
Get complete CPU set.
static hwloc_obj_t hwloc_cuda_get_device_pcidev(hwloc_topology_t topology, CUdevice cudevice)
Get the hwloc PCI device object corresponding to the CUDA device cudevice.
Definition: cuda.h:137
int hwloc_bitmap_iszero(hwloc_const_bitmap_t bitmap)
Test whether bitmap bitmap is empty.
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
int hwloc_topology_is_thissystem(hwloc_topology_t restrict topology)
Does the topology context come from this system?
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
static hwloc_obj_t hwloc_cuda_get_device_osdev(hwloc_topology_t topology, CUdevice cudevice)
Get the hwloc OS device object corresponding to CUDA device cudevice.
Definition: cuda.h:163
PCI device (filtered out by default).
Definition: hwloc.h:295
union hwloc_obj_attr_u * attr
Object type-specific Attributes, may be NULL if no attribute value was found.
Definition: hwloc.h:450
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 hwloc_obj_t hwloc_cuda_get_device_osdev_by_index(hwloc_topology_t topology, unsigned idx)
Get the hwloc OS device object corresponding to the CUDA device whose index is idx.
Definition: cuda.h:205
hwloc_obj_type_t type
Type of object.
Definition: hwloc.h:433
unsigned char dev
Device number (zz in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:677
struct hwloc_obj * parent
Parent, NULL if root (Machine object)
Definition: hwloc.h:481
static hwloc_obj_t hwloc_get_pcidev_by_busid(hwloc_topology_t topology, unsigned domain, unsigned bus, unsigned dev, unsigned func)
Find the PCI device object matching the PCI bus id given domain, bus device and function PCI bus id...
Definition: helper.h:1241
struct hwloc_topology * hwloc_topology_t
Topology context.
Definition: hwloc.h:742
unsigned char bus
Bus number (yy in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:676
char * name
Object-specific name if any. Mostly used for identifying OS devices and Misc objects where a name str...
Definition: hwloc.h:443
Operating system co-processor device. For instance "opencl0d0" for a OpenCL device, "cuda0" for a CUDA device.
Definition: hwloc.h:385
struct hwloc_obj_attr_u::hwloc_osdev_attr_s osdev
hwloc_obj_osdev_type_t type
Definition: hwloc.h:716