Hardware Locality (hwloc)  PR-737-20250925.0822.gite8f69c77f
cuda.h
1 /*
2  * SPDX-License-Identifier: BSD-3-Clause
3  * Copyright © 2010-2024 Inria. All rights reserved.
4  * Copyright © 2010-2011 Université Bordeaux
5  * Copyright © 2011 Cisco Systems, Inc. All rights reserved.
6  * See COPYING in top-level directory.
7  */
8 
17 #ifndef HWLOC_CUDA_H
18 #define HWLOC_CUDA_H
19 
20 #include "hwloc.h"
21 #include "hwloc/autogen/config.h"
22 #include "hwloc/helper.h"
23 #ifdef HWLOC_LINUX_SYS
24 #include "hwloc/linux.h"
25 #endif
26 
27 #include <cuda.h>
28 
29 
30 #ifdef __cplusplus
31 extern "C" {
32 #endif
33 
34 
50 static __hwloc_inline int
51 hwloc_cuda_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unused,
52  CUdevice cudevice, int *domain, int *bus, int *dev)
53 {
54  CUresult cres;
55 
56 #if CUDA_VERSION >= 4000
57  cres = cuDeviceGetAttribute(domain, CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID, cudevice);
58  if (cres != CUDA_SUCCESS) {
59  errno = ENOSYS;
60  return -1;
61  }
62 #else
63  *domain = 0;
64 #endif
65  cres = cuDeviceGetAttribute(bus, CU_DEVICE_ATTRIBUTE_PCI_BUS_ID, cudevice);
66  if (cres != CUDA_SUCCESS) {
67  errno = ENOSYS;
68  return -1;
69  }
70  cres = cuDeviceGetAttribute(dev, CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID, cudevice);
71  if (cres != CUDA_SUCCESS) {
72  errno = ENOSYS;
73  return -1;
74  }
75 
76  return 0;
77 }
78 
98 static __hwloc_inline int
99 hwloc_cuda_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
100  CUdevice cudevice, hwloc_cpuset_t set)
101 {
102 #ifdef HWLOC_LINUX_SYS
103  /* If we're on Linux, use the sysfs mechanism to get the local cpus */
104 #define HWLOC_CUDA_DEVICE_SYSFS_PATH_MAX 128
105  char path[HWLOC_CUDA_DEVICE_SYSFS_PATH_MAX];
106  int domainid, busid, deviceid;
107 
108  if (hwloc_cuda_get_device_pci_ids(topology, cudevice, &domainid, &busid, &deviceid))
109  return -1;
110 
111  if (!hwloc_topology_is_thissystem(topology)) {
112  errno = EINVAL;
113  return -1;
114  }
115 
116  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.0/local_cpus", domainid, busid, deviceid);
117  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
118  || hwloc_bitmap_iszero(set))
120 #else
121  /* Non-Linux systems simply get a full cpuset */
123 #endif
124  return 0;
125 }
126 
137 static __hwloc_inline hwloc_obj_t
138 hwloc_cuda_get_device_pcidev(hwloc_topology_t topology, CUdevice cudevice)
139 {
140  int domain, bus, dev;
141 
142  if (hwloc_cuda_get_device_pci_ids(topology, cudevice, &domain, &bus, &dev))
143  return NULL;
144 
145  return hwloc_get_pcidev_by_busid(topology, domain, bus, dev, 0);
146 }
147 
163 static __hwloc_inline hwloc_obj_t
164 hwloc_cuda_get_device_osdev(hwloc_topology_t topology, CUdevice cudevice)
165 {
166  hwloc_obj_t osdev = NULL;
167  int domain, bus, dev;
168 
169  if (hwloc_cuda_get_device_pci_ids(topology, cudevice, &domain, &bus, &dev))
170  return NULL;
171 
172  osdev = NULL;
173  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
174  hwloc_obj_t pcidev = osdev->parent;
175  if (strncmp(osdev->name, "cuda", 4))
176  continue;
177  if (pcidev
178  && pcidev->type == HWLOC_OBJ_PCI_DEVICE
179  && (int) pcidev->attr->pcidev.domain == domain
180  && (int) pcidev->attr->pcidev.bus == bus
181  && (int) pcidev->attr->pcidev.dev == dev
182  && pcidev->attr->pcidev.func == 0)
183  return osdev;
184  /* if PCI are filtered out, we need a info attr to match on */
185  }
186 
187  return NULL;
188 }
189 
205 static __hwloc_inline hwloc_obj_t
207 {
208  hwloc_obj_t osdev = NULL;
209  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
210  if ((osdev->attr->osdev.types & (HWLOC_OBJ_OSDEV_GPU|HWLOC_OBJ_OSDEV_COPROC)) /* assume future CUDA devices will be at least GPU or COPROC */
211  && osdev->name
212  && !strncmp("cuda", osdev->name, 4)
213  && atoi(osdev->name + 4) == (int) idx)
214  return osdev;
215  }
216  return NULL;
217 }
218 
222 #ifdef __cplusplus
223 } /* extern "C" */
224 #endif
225 
226 
227 #endif /* HWLOC_CUDA_H */
hwloc_cuda_get_device_osdev
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:164
HWLOC_OBJ_OSDEV_COPROC
@ HWLOC_OBJ_OSDEV_COPROC
Operating system co-processor device. For instance "opencl0d0" for a OpenCL device,...
Definition: hwloc.h:396
hwloc_cuda_get_device_pci_ids
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:51
hwloc_bitmap_iszero
int hwloc_bitmap_iszero(hwloc_const_bitmap_t bitmap)
Test whether bitmap bitmap is empty.
hwloc_obj
Structure of a topology object.
Definition: hwloc.h:488
HWLOC_OBJ_OSDEV_GPU
@ HWLOC_OBJ_OSDEV_GPU
Operating system GPU device. For instance ":0.0" for a GL display, "card0" for a Linux DRM device,...
Definition: hwloc.h:389
hwloc_topology_t
struct hwloc_topology * hwloc_topology_t
Topology context.
Definition: hwloc.h:778
hwloc_linux_read_path_as_cpumask
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.
hwloc_obj_attr_u::hwloc_pcidev_attr_s::dev
unsigned char dev
Device number (zz in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:727
hwloc_obj_attr_u::hwloc_pcidev_attr_s::bus
unsigned char bus
Bus number (yy in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:726
hwloc_obj::parent
struct hwloc_obj * parent
Parent, NULL if root (Machine object)
Definition: hwloc.h:538
hwloc_cuda_get_device_pcidev
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:138
hwloc_obj_attr_u::osdev
struct hwloc_obj_attr_u::hwloc_osdev_attr_s osdev
hwloc_topology_get_complete_cpuset
hwloc_const_cpuset_t hwloc_topology_get_complete_cpuset(hwloc_topology_t topology)
Get complete CPU set.
hwloc_obj::attr
union hwloc_obj_attr_u * attr
Object type-specific Attributes, may be NULL if no attribute value was found.
Definition: hwloc.h:507
hwloc_obj_attr_u::hwloc_osdev_attr_s::types
hwloc_obj_osdev_types_t types
OR'ed set of at least one hwloc_obj_osdev_type_e.
Definition: hwloc.h:761
hwloc_obj::name
char * name
Object-specific name if any. Mostly used for identifying OS devices and Misc objects where a name str...
Definition: hwloc.h:500
hwloc_cpuset_t
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:162
hwloc_obj_attr_u::hwloc_pcidev_attr_s::domain
unsigned int domain
Domain number (xxxx in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:725
HWLOC_OBJ_PCI_DEVICE
@ HWLOC_OBJ_PCI_DEVICE
PCI device (filtered out by default).
Definition: hwloc.h:320
hwloc_get_pcidev_by_busid
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:1250
hwloc_bitmap_copy
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.
hwloc_cuda_get_device_cpuset
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:99
hwloc_obj_attr_u::hwloc_pcidev_attr_s::func
unsigned char func
Function number (t in the PCI BDF notation xxxx:yy:zz.t).
Definition: hwloc.h:728
hwloc_obj_attr_u::pcidev
struct hwloc_obj_attr_u::hwloc_pcidev_attr_s pcidev
hwloc_obj::type
hwloc_obj_type_t type
Type of object.
Definition: hwloc.h:490
hwloc_get_next_osdev
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:1291
hwloc_topology_is_thissystem
int hwloc_topology_is_thissystem(hwloc_topology_t restrict topology)
Does the topology context come from this system?
hwloc_cuda_get_device_osdev_by_index
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:206