Hardware Locality (hwloc)  PR-737-20250925.0822.gite8f69c77f
cudart.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_CUDART_H
18 #define HWLOC_CUDART_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> /* for CUDA_VERSION */
28 #include <cuda_runtime_api.h>
29 
30 
31 #ifdef __cplusplus
32 extern "C" {
33 #endif
34 
35 
51 static __hwloc_inline int
52 hwloc_cudart_get_device_pci_ids(hwloc_topology_t topology __hwloc_attribute_unused,
53  int idx, int *domain, int *bus, int *dev)
54 {
55  cudaError_t cerr;
56  struct cudaDeviceProp prop;
57 
58  cerr = cudaGetDeviceProperties(&prop, idx);
59  if (cerr) {
60  errno = ENOSYS;
61  return -1;
62  }
63 
64 #if CUDA_VERSION >= 4000
65  *domain = prop.pciDomainID;
66 #else
67  *domain = 0;
68 #endif
69 
70  *bus = prop.pciBusID;
71  *dev = prop.pciDeviceID;
72 
73  return 0;
74 }
75 
95 static __hwloc_inline int
96 hwloc_cudart_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
97  int idx, hwloc_cpuset_t set)
98 {
99 #ifdef HWLOC_LINUX_SYS
100  /* If we're on Linux, use the sysfs mechanism to get the local cpus */
101 #define HWLOC_CUDART_DEVICE_SYSFS_PATH_MAX 128
102  char path[HWLOC_CUDART_DEVICE_SYSFS_PATH_MAX];
103  int domain, bus, dev;
104 
105  if (hwloc_cudart_get_device_pci_ids(topology, idx, &domain, &bus, &dev))
106  return -1;
107 
108  if (!hwloc_topology_is_thissystem(topology)) {
109  errno = EINVAL;
110  return -1;
111  }
112 
113  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.0/local_cpus", (unsigned) domain, (unsigned) bus, (unsigned) dev);
114  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
115  || hwloc_bitmap_iszero(set))
117 #else
118  /* Non-Linux systems simply get a full cpuset */
120 #endif
121  return 0;
122 }
123 
134 static __hwloc_inline hwloc_obj_t
136 {
137  int domain, bus, dev;
138 
139  if (hwloc_cudart_get_device_pci_ids(topology, idx, &domain, &bus, &dev))
140  return NULL;
141 
142  return hwloc_get_pcidev_by_busid(topology, domain, bus, dev, 0);
143 }
144 
162 static __hwloc_inline hwloc_obj_t
164 {
165  hwloc_obj_t osdev = NULL;
166  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
167  if ((osdev->attr->osdev.types & (HWLOC_OBJ_OSDEV_GPU|HWLOC_OBJ_OSDEV_COPROC)) /* assume future CUDA devices will be at least GPU or COPROC */
168  && osdev->name
169  && !strncmp("cuda", osdev->name, 4)
170  && atoi(osdev->name + 4) == (int) idx)
171  return osdev;
172  }
173  return NULL;
174 }
175 
179 #ifdef __cplusplus
180 } /* extern "C" */
181 #endif
182 
183 
184 #endif /* HWLOC_CUDART_H */
hwloc_cudart_get_device_cpuset
int hwloc_cudart_get_device_cpuset(hwloc_topology_t topology, int idx, hwloc_cpuset_t set)
Get the CPU set of processors that are physically close to device idx.
Definition: cudart.h:96
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_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_cudart_get_device_pci_ids
int hwloc_cudart_get_device_pci_ids(hwloc_topology_t topology, int idx, int *domain, int *bus, int *dev)
Return the domain, bus and device IDs of the CUDA device whose index is idx.
Definition: cudart.h:52
hwloc_cudart_get_device_osdev_by_index
hwloc_obj_t hwloc_cudart_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: cudart.h:163
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_cudart_get_device_pcidev
hwloc_obj_t hwloc_cudart_get_device_pcidev(hwloc_topology_t topology, int idx)
Get the hwloc PCI device object corresponding to the CUDA device whose index is idx.
Definition: cudart.h:135
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_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?