Open MPI logo

Portable Hardware Locality (hwloc) Documentation: v2.8.0

  |   Home   |   Support   |   FAQ   |  
cuda.h
1 /*
2  * Copyright © 2010-2021 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 
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 
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))
113 #else
114  /* Non-Linux systems simply get a full cpuset */
116 #endif
117  return 0;
118 }
119 
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 
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 
198 static __hwloc_inline hwloc_obj_t
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 
215 #ifdef __cplusplus
216 } /* extern "C" */
217 #endif
218 
219 
220 #endif /* HWLOC_CUDA_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:1194
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:1157
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_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:47
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:131
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:199
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:157
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:92
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