Open MPI logo

Portable Hardware Locality (hwloc) Documentation: v2.2.0

  |   Home   |   Support   |   FAQ   |  
opencl.h
1 /*
2  * Copyright ¬© 2012-2019 Inria. All rights reserved.
3  * Copyright ¬© 2013, 2018 Universit√© 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 
73 static __hwloc_inline int
75  unsigned *domain, unsigned *bus, unsigned *dev, unsigned *func)
76 {
78  cl_uint nvbus, nvslot, nvdomain;
79  cl_int clret;
80 
81  clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_TOPOLOGY_AMD, sizeof(amdtopo), &amdtopo, NULL);
82  if (CL_SUCCESS == clret
83  && HWLOC_CL_DEVICE_TOPOLOGY_TYPE_PCIE_AMD == amdtopo.raw.type) {
84  *domain = 0; /* can't do anything better */
85  *bus = (unsigned) amdtopo.pcie.bus;
86  *dev = (unsigned) amdtopo.pcie.device;
87  *func = (unsigned) amdtopo.pcie.function;
88  return 0;
89  }
90 
91  clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_BUS_ID_NV, sizeof(nvbus), &nvbus, NULL);
92  if (CL_SUCCESS == clret) {
93  clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_SLOT_ID_NV, sizeof(nvslot), &nvslot, NULL);
94  if (CL_SUCCESS == clret) {
95  clret = clGetDeviceInfo(device, HWLOC_CL_DEVICE_PCI_DOMAIN_ID_NV, sizeof(nvdomain), &nvdomain, NULL);
96  if (CL_SUCCESS == clret) { /* available since CUDA 10.2 */
97  *domain = nvdomain;
98  } else {
99  *domain = 0;
100  }
101  *bus = nvbus & 0xff;
102  /* non-documented but used in many other projects */
103  *dev = nvslot >> 3;
104  *func = nvslot & 0x7;
105  return 0;
106  }
107  }
108 
109  return -1;
110 }
111 
129 static __hwloc_inline int
130 hwloc_opencl_get_device_cpuset(hwloc_topology_t topology __hwloc_attribute_unused,
131  cl_device_id device __hwloc_attribute_unused,
132  hwloc_cpuset_t set)
133 {
134 #if (defined HWLOC_LINUX_SYS)
135  /* If we're on Linux, try AMD/NVIDIA extensions + the sysfs mechanism to get the local cpus */
136 #define HWLOC_OPENCL_DEVICE_SYSFS_PATH_MAX 128
137  char path[HWLOC_OPENCL_DEVICE_SYSFS_PATH_MAX];
138  unsigned pcidomain, pcibus, pcidev, pcifunc;
139 
140  if (!hwloc_topology_is_thissystem(topology)) {
141  errno = EINVAL;
142  return -1;
143  }
144 
145  if (hwloc_opencl_get_device_pci_busid(device, &pcidomain, &pcibus, &pcidev, &pcifunc) < 0) {
147  return 0;
148  }
149 
150  sprintf(path, "/sys/bus/pci/devices/%04x:%02x:%02x.%01x/local_cpus", pcidomain, pcibus, pcidev, pcifunc);
151  if (hwloc_linux_read_path_as_cpumask(path, set) < 0
152  || hwloc_bitmap_iszero(set))
154 #else
155  /* Non-Linux systems simply get a full cpuset */
157 #endif
158  return 0;
159 }
160 
176 static __hwloc_inline hwloc_obj_t
178  unsigned platform_index, unsigned device_index)
179 {
180  unsigned x = (unsigned) -1, y = (unsigned) -1;
181  hwloc_obj_t osdev = NULL;
182  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
183  if (HWLOC_OBJ_OSDEV_COPROC == osdev->attr->osdev.type
184  && osdev->name
185  && sscanf(osdev->name, "opencl%ud%u", &x, &y) == 2
186  && platform_index == x && device_index == y)
187  return osdev;
188  }
189  return NULL;
190 }
191 
212 static __hwloc_inline hwloc_obj_t
213 hwloc_opencl_get_device_osdev(hwloc_topology_t topology __hwloc_attribute_unused,
214  cl_device_id device __hwloc_attribute_unused)
215 {
216  hwloc_obj_t osdev;
217  unsigned pcidomain, pcibus, pcidevice, pcifunc;
218 
219  if (hwloc_opencl_get_device_pci_busid(device, &pcidomain, &pcibus, &pcidevice, &pcifunc) < 0) {
220  errno = EINVAL;
221  return NULL;
222  }
223 
224  osdev = NULL;
225  while ((osdev = hwloc_get_next_osdev(topology, osdev)) != NULL) {
226  hwloc_obj_t pcidev = osdev->parent;
227  if (strncmp(osdev->name, "opencl", 6))
228  continue;
229  if (pcidev
230  && pcidev->type == HWLOC_OBJ_PCI_DEVICE
231  && pcidev->attr->pcidev.domain == pcidomain
232  && pcidev->attr->pcidev.bus == pcibus
233  && pcidev->attr->pcidev.dev == pcidevice
234  && pcidev->attr->pcidev.func == pcifunc)
235  return osdev;
236  /* if PCI are filtered out, we need a info attr to match on */
237  }
238 
239  return NULL;
240 }
241 
245 #ifdef __cplusplus
246 } /* extern "C" */
247 #endif
248 
249 
250 #endif /* HWLOC_OPENCL_H */
static int hwloc_opencl_get_device_cpuset(hwloc_topology_t topology, cl_device_id device, hwloc_cpuset_t set)
Get the CPU set of logical processors that are physically close to OpenCL device device.
Definition: opencl.h:130
Definition: opencl.h:46
cl_char bus
Definition: opencl.h:48
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:1151
hwloc_obj_type_t type
Type of object.
Definition: hwloc.h:386
struct hwloc_obj_attr_u::hwloc_osdev_attr_s osdev
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:213
struct hwloc_obj_attr_u::hwloc_pcidev_attr_s pcidev
cl_char device
Definition: opencl.h:48
struct hwloc_topology * hwloc_topology_t
Topology context.
Definition: hwloc.h:678
cl_char function
Definition: opencl.h:48
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:74
@ HWLOC_OBJ_OSDEV_COPROC
Operating system co-processor device. For instance "mic0" for a Xeon Phi (MIC) on Linux,...
Definition: hwloc.h:340
unsigned char bus
Definition: hwloc.h:625
union hwloc_obj_attr_u * attr
Object type-specific Attributes, may be NULL if no attribute value was found.
Definition: hwloc.h:403
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
cl_uint type
Definition: opencl.h:47
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.
Structure of a topology object.
Definition: hwloc.h:384
struct hwloc_cl_device_topology_amd::@4 pcie
int hwloc_bitmap_iszero(hwloc_const_bitmap_t bitmap)
Test whether bitmap bitmap is empty.
unsigned char dev
Definition: hwloc.h:625
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:177
int hwloc_topology_is_thissystem(hwloc_topology_t restrict topology)
Does the topology context come from this system?
@ HWLOC_OBJ_PCI_DEVICE
PCI device (filtered out by default). They are not added to the topology unless I/O discovery is enab...
Definition: hwloc.h:266
char * name
Object-specific name if any. Mostly used for identifying OS devices and Misc objects where a name str...
Definition: hwloc.h:396
hwloc_obj_osdev_type_t type
Definition: hwloc.h:652
struct hwloc_cl_device_topology_amd::@3 raw
unsigned char func
Definition: hwloc.h:625
unsigned short domain
Definition: hwloc.h:621
struct hwloc_obj * parent
Parent, NULL if root (Machine object)
Definition: hwloc.h:434
hwloc_const_cpuset_t hwloc_topology_get_complete_cpuset(hwloc_topology_t topology)
Get complete CPU set.