1CLINFO(1) General Commands Manual CLINFO(1)
2
6 clinfo - show OpenCL platforms and devices
7
10 clinfo [options ...]
11
14 clinfo prints all available information about all OpenCL platforms
15 available on the system and the devices they expose.
16
19 clinfo accepts the following options:
20 -a, --all-props
22 try to retrieve all properties, even those not officially supported
23 (e.g. because they require specific extensions), but only show them
24 if the property could be retrieved successfully; see also the LIMITA‐
25 TIONS section below; note that even though this may reveal hidden
26 properties, there is no guarantee that the returned values are mean‐
27 ingful, nor that the corresponding feature is actually available at
28 all;
29 -A, --always-all-props
31 like -a, but also show errors;
32 --human
34 produce human-friendly output; this is the default (except as noted
35 below);
36 --raw
38 produce machine-friendly output; this is the default if clinfo is in‐
39 voked with a name that contains the string “raw”;
40 --json
42 outputs the raw data (cf. the --raw option) in JSON format; support
43 for this option is experimental, as the representation of some of the
44 values is not finalized;
45 --offline
47 shows also offline devices for platforms that expose this feature;
48 --null-platform
50 tries to handle the NULL platform as a normal platform, retrieving
51 and showing its properties and devices; this is in addition to the
52 NULL platform behavior tests done at the end, and can be useful on
53 systems where there are no ICD platforms, but there is a platform
54 hard-coded in the OpenCL library itself;
55 -l, --list
57 list platforms and devices by name, with no (other) properties;
58 -dplatform_index:device_index
60 --device platform_index:device_index
61 only show properties for the specified device in the specified plat‐
62 form;
63 --prop property-name
65 only show properties whose symbolic name matches (contains as a sub‐
66 string) the given property-name; the name is normalized as upper-case
67 and with minus sign (-) replaced by underscore signs (_); when this
68 flag is specified, raw mode is forced;
69 -h, -?
71 show usage;
72 --version, -v
74 show program version.
75
78 OpenCL 1.1, OpenCL 1.2, OpenCL 2.0, OpenCL 2.1, OpenCL 2.2, OpenCL 3.0.
79
82 Supported OpenCL extensions:
83 cl_khr_device_uuid
85 for the UUID, LUID and node mask of the device;
86 cl_khr_extended_versioning
88 for the extended platform, device, extension and IL versioned proper‐
89 ties backported from OpenCL 3.0 to previous OpenCL versions;
90 cl_khr_fp16, cl_khr_fp64, cl_amd_fp64, cl_APPLE_fp64_basic_ops
92 for information about support for half-precision and double-precision
93 floating-point data types;
94 cl_khr_image2d_from_buffer
96 for information about the base address and pitch alignment require‐
97 ments of buffers to be used as base for 2D images;
98 cl_khr_il_program
100 for information about the supported IL (Intermediate Language) repre‐
101 sentations;
102 cl_khr_spir
104 for information about the supported SPIR (Standard Portable Interme‐
105 diate Representation) versions;
106 cl_khr_icd
108 for the suffix of vendor extensions functions;
109 cl_khr_subgroup_named_barrier
111 for the maximum number of named sub-group barriers;
112 cl_khr_terminate_context, cl_arm_controlled_kernel_termination
114 for the terminate capabilities for the device;
115 cl_ext_device_fission
117 for device fission support in OpenCL 1.1 devices;
118 cl_khr_pci_bus_info
120 for the PCI bus information (see also cl_nv_device_attribute_query
121 and cl_amd_device_attribute_query)
122 cl_ext_atomic_counters_32
124 cl_ext_atomic_counters_64
125 for the atomic counter extension;
126 cl_ext_cxx_for_opencl
128 for the version of the C++ for OpenCL language supported by the de‐
129 vice compiler;
130 cl_amd_device_attribute_query
132 for AMD-specific device attributes;
133 cl_amd_object_metadata
135 to show the maximum number of keys supported by the platform;
136 cl_amd_offline_devices
138 to show offline devices exposed by the platform, if requested (see
139 --offline option);
140 cl_amd_copy_buffer_p2p
142 to show the number and IDs of available P2P devices;
143 cl_amd_svm
145 cl_arm_shared_virtual_memory
146 for Shared Virtual Memory (SVM) capabilities in OpenCL 1.2 devices;
147 cl_arm_core_id
149 to show the (potentially sparse) list of the core IDs that the device
150 may return;
151 cl_arm_job_slot_selection
153 to show the (potentially sparse) list of available job slots for com‐
154 mand submission;
155 cl_arm_scheduling_controls
157 to show the supported work scheduling controls and the available sets
158 of register allocations;
159 cl_nv_device_attribute_query
161 for NVIDIA-specific device attributes;
162 cl_intel_device_attribute_query
164 for Intel-specific device attributes;
165 cl_intel_exec_by_local_thread
167 for the Intel extension allowing CPU devices to run kernels as part
168 of the current host thread;
169 cl_intel_advanced_motion_estimation
171 for the version of the Intel Motion Estimation accelerator version;
172 cl_intel_device_side_avc_motion_estimation
174 for the version and supported features of Intel's device-side AVC Mo‐
175 tion;
176 cl_intel_planar_yuv
178 for the maximum dimensions of planar YUV images;
179 cl_intel_simultaneous_sharing
181 for simultaneous CL/GL/DirectX context sharing (only partial sup‐
182 port);
183 cl_intel_required_subgroup_size
185 to enumerate allowed sub-group sizes;
186 cl_intel_command_queue_families
188 to enumerate the available command queues and their properties and
189 capabilities;
190 cl_altera_device_temperature
192 for the Altera extension to query the core temperature of the device;
193 cl_qcom_ext_host_ptr
195 for the QUALCOMM extension to query page size and required padding in
196 external memory allocation.
197
200 Some information is duplicated when available from multiple sources.
201 Examples:
202 • supported device partition types and domains as obtained using the
204 cl_ext_device_fission extension typically match the ones obtained us‐
205 ing the core OpenCL 1.2 device partition feature;
206 • the preferred work-group size multiple matches the NVIDIA warp size
208 (on NVIDIA devices) or the AMD wavefront width (on AMD devices).
209 Some floating-point configuration flags may only be meaningful for spe‐
212 cific precisions and/or specific OpenCL versions. For example,
213 CL_FP_CORRECTLY_ROUNDED_DIVIDE_SQRT is only relevant for single preci‐
214 sion in OpenCL 1.2 devices.
215 The implementation-defined behavior for NULL platform or context prop‐
218 erties is tested for the following API calls:
219 clGetPlatformInfo()
221 by trying to show the platform name;
222 clGetDeviceIDs()
224 by trying to enumerate devices; the corresponding platform (if any)
225 is then detected by querying the device platform of the first device;
226 clCreateteContext()
228 by trying to create a context from a device from the previous list
229 (if any), and a context from a device from a different platform;
230 clCreateteContextFromType()
232 by trying to create contexts for each device type (except DEFAULT).
233
236 Support for OpenCL 2.x properties is not fully tested.
237 Support for cl_khr_subgroup_named_barrier is experimental due to miss‐
240 ing definitions in the official OpenCL headers.
241 Raw (machine-parsable) output is considered experimental, the output
244 format might still undergo changes.
245 The properties of the ICD loader will also be queried if the clGetICD‐
248 LoaderInfoOCLICD extension function is found.
249 Support for the properties exposed by cl_amd_copy_buffer_p2p is experi‐
252 mental.
253 Support for some (documented and undocumented) properties exposed by
256 cl_amd_device_attribute_query is experimental (see also LIMITATIONS).
257 Support for the interop lists exposed by cl_intel_simultaneous_sharing
260 is experimental.
261 The highest OpenCL version supported by the ICD loader is detected with
264 some trivial heuristics (symbols found); a notice is output if this is
265 lower than the highest platform OpenCL version, or if the detected ver‐
266 sion doesn't match the one declared by the ICD loader itself.
267
270 OpenCL provides no explicit mean to detect the supported version of any
271 extension exposed by a device, which makes it impossible to determine a
272 priori if it will be possible to successfully query a device about a
273 specific property. Additionally, the actual size and meaning of some
274 properties are not officially declared anywhere.
275 Most notably, this affects extensions such as cl_amd_device_attri‐
278 bute_query, cl_nv_device_attribute_query and cl_arm_core_id. Heuris‐
279 tics based on standard version support are partially used in the code
280 to determine which version may be supported.
281 Properties which are known to be affected by these limitations include:
284 CL_DEVICE_GLOBAL_FREE_MEMORY_AMD
287 documented in v3 of the cl_amd_device_attribute_query extension spec‐
288 ification as being the global free memory in KBytes, without any ex‐
289 planation given on why there are two values, although in the source
290 code of the ROCm stack the second value is documented as being the
291 largest free block;
292 CL_DEVICE_AVAILABLE_ASYNC_QUEUES_AMD
294 documented in v3 of the cl_amd_device_attribute_query extension spec‐
295 ification, but not reported by drivers supporting other v3 proper‐
296 ties. This has now been enabled for drivers assumed to support v4 of
297 the same extension;
298 CL_DEVICE_TERMINATE_CAPABILITY_KHR
300 exposed by the cl_khr_terminate_context has changed value between
301 OpenCL 1.x and 2.x, and it's allegedly a bitfield, whose values are
302 however not defined anywhere.
303
306 General
307 Please report any issues on the project tracker on GitHub
308 ⟨http://github.com/Oblomov/clinfo⟩.
309 LLVM CommandLine errors
312 If multiple OpenCL platforms using shared LLVM libraries are present in
313 the system, clinfo (and other OpenCL application) may crash with errors
314 to the tune of
315 : CommandLine Error: Option '(some option name)' registered more than once!
317 LLVM ERROR: inconsistency in registered CommandLine options
318 or similar. This is not an issue in clinfo, or in any OpenCL platform
320 or application, but it is due to the way LLVM handles its own command-
321 line options parsing. The issue has been reported upstream as issue
322 #30587 ⟨https://bugs.llvm.org/show_bug.cgi?id=30587⟩. See the next
323 point for possible workarounds and assistance in identifying the con‐
324 flicting platforms.
325 Segmentation faults
328 Faulty OpenCL platforms may cause segmentation faults in clinfo during
329 the information gathering phase, sometimes even before any output is
330 shown. There is very little clinfo can do to avoid this. If you see
331 this happening, try disabling all platforms and then re-enabling them
332 one by one until you experience the crash again. Chances are the last
333 platform you enabled is defective in some way (either by being incom‐
334 patible with other platforms or by missing necessary components and not
335 handling their absence gracefully).
336 To selectively enable/disable platforms, one way is to move or rename
339 the *.icd files present in /etc/OpenCL/vendors/ and then restoring them
340 one by one. When using the free-software ocl-icd OpenCL library, a sim‐
341 ilar effect can be achieved by setting the OPENCL_VENDOR_PATH or
342 OCL_ICD_VENDORS environment variables, as documented in libOpenCL(7).
343 Other implementations of libOpenCL are known to support OPENCL_VEN‐
344 DOR_PATH too.
345 Example
348 find /etc/OpenCL/vendors/ -name '*.icd' | while read OPENCL_VEN‐
349 DOR_PATH ; do clinfo -l > /dev/null ; echo "$? ${OPENCL_VENDOR_PATH}"
350 ; done
351 This one liner will run clinfo -l for each platform individually (hid‐
354 ing the normal output), and report the .icd path prefixed by 0 for suc‐
355 cessful runs, and a non-zero value for faulty platforms.
356clinfo 3.0.23.01.25 2023-01-25 CLINFO(1)