Brice Videau [Tue, 17 Nov 2015 10:32:32 +0000 (11:32 +0100)]
Added missing clGetKernelSubGroupInfoKHR.
Brice Videau [Tue, 17 Nov 2015 10:31:50 +0000 (11:31 +0100)]
Added missing EGL entries.
Vincent Danjean [Tue, 9 Jun 2015 10:01:48 +0000 (12:01 +0200)]
cleanup doc
Vincent Danjean [Tue, 9 Jun 2015 09:46:53 +0000 (11:46 +0200)]
Explicitly set the revision date in the documentation
This allows to always produce the same doc, whatever localtime is.
Note that the 'footer-style' attribute (used when creating the HTML page)
is a patch of asciidoc available in Debian but not merged upstream yet.
Vincent Danjean [Mon, 8 Jun 2015 23:14:08 +0000 (01:14 +0200)]
New release
Vincent Danjean [Mon, 8 Jun 2015 23:12:56 +0000 (01:12 +0200)]
Also use real prototype in weak test functions
Vincent Danjean [Mon, 8 Jun 2015 18:30:34 +0000 (20:30 +0200)]
Release 2.2.6
Vincent Danjean [Mon, 8 Jun 2015 11:19:50 +0000 (13:19 +0200)]
Improve documentation (OCL_ICD_PLATFORM_SORT and OCL_ICD_DEFAULT_PLATFORM)
Vincent Danjean [Mon, 8 Jun 2015 11:18:37 +0000 (13:18 +0200)]
allows to avoid the platform sort and extends the testsuite
Vincent Danjean [Sun, 7 Jun 2015 23:03:22 +0000 (01:03 +0200)]
Fix and improve testsuite
Run through our ICD loader (or the system ICD Loader when updating
the database) is now using the whole prototypes of OpenCL functions.
And, when testing our ICD Loader, the result is matched against what
is expected.
Vincent Danjean [Sat, 6 Jun 2015 16:19:28 +0000 (18:19 +0200)]
Release 2.2.5
Vincent Danjean [Sat, 6 Jun 2015 16:05:23 +0000 (18:05 +0200)]
[BUG] call the function when the return type is void
Before this patch, clSVMFree was never called...
Vincent Danjean [Wed, 3 Jun 2015 16:00:08 +0000 (18:00 +0200)]
Correctly handle clGetDeviceIDs in generator and tests
As this function is used by our implementation (to get
the number of supported GPU/CPU/Devices), it has to be
handled specifically
Vincent Danjean [Wed, 3 Jun 2015 12:45:18 +0000 (14:45 +0200)]
Sort platforms in one pass
Use qsort from the C library.
Moreover, the code will be easier to modify
when/if the sort will be done with external/different criteria
(admin or user preferences, etc.)
Giuseppe Bilotta [Sun, 16 Nov 2014 13:48:31 +0000 (14:48 +0100)]
Sort platforms by number of GPU, CPU and ALL devices
Giuseppe Bilotta [Sun, 16 Nov 2014 13:03:35 +0000 (14:03 +0100)]
Count GPU, CPU and ALL devices on newly inserted platforms
Giuseppe Bilotta [Sun, 16 Nov 2014 12:29:45 +0000 (13:29 +0100)]
Remove useless vendor_icd.first_platform
Giuseppe Bilotta [Sat, 4 Apr 2015 06:19:05 +0000 (08:19 +0200)]
Whitespace cleanup
Giuseppe Bilotta [Thu, 13 Nov 2014 17:34:41 +0000 (18:34 +0100)]
Fix a few typos
Vincent Danjean [Mon, 18 May 2015 14:42:09 +0000 (16:42 +0200)]
Release 2.2.4
Vincent Danjean [Mon, 18 May 2015 14:40:54 +0000 (16:40 +0200)]
Fix initialization in case of default platform use
Vincent Danjean [Mon, 18 May 2015 14:31:35 +0000 (16:31 +0200)]
Improve tests
Vincent Danjean [Mon, 18 May 2015 14:31:18 +0000 (16:31 +0200)]
avoid loading ICD at library init time
this avoid recursive dlopen
see https://bugzilla.redhat.com/show_bug.cgi?id=
1219646
Vincent Danjean [Mon, 18 May 2015 14:26:59 +0000 (16:26 +0200)]
Avoid using system-wide vendors files at test time
Brice Videau [Mon, 6 Oct 2014 07:10:09 +0000 (09:10 +0200)]
Consistency.
Vincent Danjean [Sun, 5 Oct 2014 15:04:26 +0000 (17:04 +0200)]
Release 2.2.3
Vincent Danjean [Sun, 5 Oct 2014 15:02:57 +0000 (17:02 +0200)]
Fix numbers in unknown members of dispatch table to avoid to reuse one
Vincent Danjean [Sun, 5 Oct 2014 09:08:00 +0000 (11:08 +0200)]
Release 2.2.2
Vincent Danjean [Sun, 5 Oct 2014 09:07:08 +0000 (11:07 +0200)]
Avoid using types not supported by OpenCL headers in ocl_icd.h
If 1.2 headers are installed, only ICD up to 1.2 can be created.
But no type error will be generated when using ocl_icd.h.
Vincent Danjean [Sun, 5 Oct 2014 09:03:25 +0000 (11:03 +0200)]
Avoid warnings with 2.0 OpenCL headers when using 1.2 deprecated functions
Vincent Danjean [Wed, 1 Oct 2014 12:07:27 +0000 (14:07 +0200)]
Release 2.2.1
Vincent Danjean [Wed, 1 Oct 2014 12:06:19 +0000 (14:06 +0200)]
add official Khronos OpenCL headers into repo
- remove unofficial Debian OpenCL headers package
- add support in the configure script to use these headers
Vincent Danjean [Fri, 26 Sep 2014 14:09:43 +0000 (16:09 +0200)]
Release 2.2.0
Vincent Danjean [Fri, 26 Sep 2014 14:06:34 +0000 (16:06 +0200)]
Bump versions
Vincent Danjean [Fri, 26 Sep 2014 14:06:25 +0000 (16:06 +0200)]
Bug fix in code path rarely used
Vincent Danjean [Fri, 26 Sep 2014 12:33:49 +0000 (14:33 +0200)]
Interface 2.0
Correct order taken from the AMD implementation (fglrx-14.41)
Aaron Watry [Mon, 1 Sep 2014 14:15:06 +0000 (16:15 +0200)]
Don't read past the potential end of the lib_path string
Aaron Watry [Mon, 1 Sep 2014 14:15:50 +0000 (16:15 +0200)]
Fix directory resource leak
Brice Videau [Thu, 22 May 2014 11:45:48 +0000 (13:45 +0200)]
Commented 2.0 OpenCL API.
Brice Videau [Thu, 22 May 2014 11:44:47 +0000 (13:44 +0200)]
Added up to date headers package.
Brice Videau [Thu, 22 May 2014 09:33:37 +0000 (11:33 +0200)]
Generator can process 2.0 functions.
Brice Videau [Wed, 21 May 2014 13:03:14 +0000 (15:03 +0200)]
Added OpenCL 2.0 functions. Hopefully at the right position (to confirm).
Vincent Danjean [Tue, 1 Apr 2014 07:36:36 +0000 (09:36 +0200)]
Remove -Werror for AM_CPPFLAGS by default
Vincent Danjean [Sun, 22 Dec 2013 11:19:30 +0000 (12:19 +0100)]
Release 2.1.3
Vincent Danjean [Tue, 3 Dec 2013 00:10:59 +0000 (01:10 +0100)]
Refactor code for selection of default platform
The default platform is calculated only once in a run
This can lead to a bug if the OPENCL_ICD_DEFAULT_PLATFORM envvar is modified during the execution of the program
Vincent Danjean [Tue, 3 Dec 2013 00:07:57 +0000 (01:07 +0100)]
[build] add tests for default platform selection
Vincent Danjean [Mon, 2 Dec 2013 23:42:09 +0000 (00:42 +0100)]
All generated file ends with "_gen" suffix (but installed once)
Only ocl_icd.h and ocl_icd_bindings.c are generated sources without _gen suffix
Vincent Danjean [Mon, 2 Dec 2013 12:52:09 +0000 (13:52 +0100)]
Rewrote initialization comments
Brice Videau [Wed, 20 Nov 2013 01:28:35 +0000 (18:28 -0700)]
In case a NULL platform is passed to the loader, the default platform is selected if it exists and is valid.
Brice Videau [Wed, 20 Nov 2013 01:27:38 +0000 (18:27 -0700)]
If we have no valid platforms non can be valid.
Brice Videau [Wed, 20 Nov 2013 01:27:01 +0000 (18:27 -0700)]
Moved some declarations to the header as they are needed in the generated part now.
Vincent Danjean [Wed, 23 Oct 2013 10:18:18 +0000 (12:18 +0200)]
Release 2.1.2
Vincent Danjean [Wed, 23 Oct 2013 10:25:47 +0000 (12:25 +0200)]
Also check for NULL events
Vincent Danjean [Wed, 23 Oct 2013 10:14:37 +0000 (12:14 +0200)]
Refactor code and compilation fix in clGetGLContextInfoKHR
Brice Videau [Wed, 23 Oct 2013 08:24:33 +0000 (10:24 +0200)]
Added platform checking for clGetGLContextInfoKHR.
Brice Videau [Wed, 23 Oct 2013 08:17:08 +0000 (10:17 +0200)]
Code cleanup.
Brice Videau [Wed, 23 Oct 2013 08:13:44 +0000 (10:13 +0200)]
cl_platform_id s provided by users to clCreateContext and clCreateContextFromType will be checked against the list of platforms gathered during setup to assert validity.
Brice Videau [Wed, 23 Oct 2013 07:54:07 +0000 (09:54 +0200)]
Check for platform validity. Could be improved.
Brice Videau [Wed, 23 Oct 2013 07:53:14 +0000 (09:53 +0200)]
Indentation fix
Vincent Danjean [Tue, 22 Oct 2013 19:19:50 +0000 (21:19 +0200)]
Release 2.1.1
Vincent Danjean [Tue, 22 Oct 2013 19:19:13 +0000 (21:19 +0200)]
Fix missing NULL argument handling in clCreateContext
Vincent Danjean [Wed, 2 Oct 2013 12:33:48 +0000 (14:33 +0200)]
Release 2.1.0
Vincent Danjean [Wed, 2 Oct 2013 12:33:09 +0000 (14:33 +0200)]
Refactor generated code
Brice Videau [Tue, 1 Oct 2013 09:04:51 +0000 (11:04 +0200)]
Added NULL pointers handling, should not segfault anymore on NULL pointer.
Vincent Danjean [Wed, 24 Jul 2013 08:53:17 +0000 (10:53 +0200)]
Release 2.0.4
Vincent Danjean [Wed, 24 Jul 2013 08:52:32 +0000 (10:52 +0200)]
By default, show all compile commands
Vincent Danjean [Sat, 22 Jun 2013 20:10:36 +0000 (22:10 +0200)]
Fix configure.ac for automake 1.13
AM_PROG_AR and AM_PROC_CC_C_O are both required with automake 1.13
when running autoreconf -vif
Brice Videau [Thu, 20 Jun 2013 09:31:04 +0000 (11:31 +0200)]
Include guard for ocl_icd.h
Vincent Danjean [Tue, 28 May 2013 19:18:56 +0000 (21:18 +0200)]
Release 2.0.3
Update NEWS file
Vincent Danjean [Tue, 28 May 2013 19:16:27 +0000 (21:16 +0200)]
Fix PATH for "make installcheck"
use absolute path to be sure to point to the good directory.
Brice Videau [Thu, 28 Mar 2013 09:57:21 +0000 (10:57 +0100)]
Small potential bugfix and hopefull norm interpretation. If asked the number of platform returned will be 0 in case of CL_PLATFORM_NOT_FOUND_KHR.
Vincent Danjean [Tue, 12 Feb 2013 10:31:04 +0000 (11:31 +0100)]
Update NEWS file
Vincent Danjean [Mon, 11 Feb 2013 15:31:26 +0000 (16:31 +0100)]
icd_generator: still cleanup
Vincent Danjean [Mon, 11 Feb 2013 14:10:43 +0000 (15:10 +0100)]
refactor icd_generator
* move and briefly comment methods groups
Vincent Danjean [Mon, 11 Feb 2013 13:54:46 +0000 (14:54 +0100)]
cleanup icd_generator
* rename options of icd_generator
* ex 'finalize' mode (now 'update-database') does not run run_dummy_icd itself
The output is passed as a file option instead
Vincent Danjean [Mon, 11 Feb 2013 12:50:29 +0000 (13:50 +0100)]
Merge branch 'master' of git+ssh://scm.forge.imag.fr/scmrepos/git/ocl-icd/ocl-icd
Vincent Danjean [Mon, 11 Feb 2013 12:50:20 +0000 (13:50 +0100)]
Add clTerminateContextKHR in testsuite
Brice Videau [Wed, 16 Jan 2013 11:20:33 +0000 (12:20 +0100)]
New version
Brice Videau [Wed, 16 Jan 2013 11:20:25 +0000 (12:20 +0100)]
fix bug when loading a correct icd *after* rejecting one
Vincent Danjean [Wed, 5 Dec 2012 12:01:47 +0000 (13:01 +0100)]
Default platform is now specified
Perhaps this should be generalized to other functions
From the doc:
3: How will the ICD handle a NULL cl_platform_id?
RESOLVED: The ICD will by default choose the first enumerated platform as the NULL
platform. The user can override this default by setting an environment variable
OPENCL_ICD_DEFAULT_PLATFORM to the desired platform index. The API calls that deal
with platforms will return CL_INVALID_PLATFORM if the index is not between zero and
(number of platforms - 1), both inclusive.
Vincent Danjean [Wed, 5 Dec 2012 11:26:00 +0000 (12:26 +0100)]
[build] lower requirement on autoconf
2.67 seems enough and this is the version in Debian squeeze
Brice Videau [Tue, 4 Dec 2012 16:23:15 +0000 (17:23 +0100)]
Added windows specific functions to the table, then prevented their use in our loader for now.
Vincent Danjean [Fri, 5 Oct 2012 12:25:47 +0000 (14:25 +0200)]
Implement clGetICDLoaderInfoOCLICD extension with a test case
Vincent Danjean [Fri, 5 Oct 2012 11:49:18 +0000 (13:49 +0200)]
cleanup code
Vincent Danjean [Fri, 5 Oct 2012 11:48:55 +0000 (13:48 +0200)]
clGetExtensionFunctionAddress should return only extensions, not core functions
Vincent Danjean [Fri, 5 Oct 2012 11:46:27 +0000 (13:46 +0200)]
Fix bug for dumping the internal ICD structure
Vincent Danjean [Fri, 14 Sep 2012 09:10:13 +0000 (11:10 +0200)]
Merge branch 'master'
Vincent Danjean [Fri, 14 Sep 2012 08:52:24 +0000 (10:52 +0200)]
New release
Vincent Danjean [Fri, 14 Sep 2012 08:44:53 +0000 (10:44 +0200)]
New release
Vincent Danjean [Fri, 14 Sep 2012 08:41:50 +0000 (10:41 +0200)]
[doc] fix another typo found by lintian
Vincent Danjean [Fri, 14 Sep 2012 08:41:33 +0000 (10:41 +0200)]
[doc] remove unused synopsis
Vincent Danjean [Fri, 14 Sep 2012 07:55:23 +0000 (09:55 +0200)]
[doc] fix typo found by lintian
Vincent Danjean [Thu, 13 Sep 2012 15:16:08 +0000 (17:16 +0200)]
New release
Dump version to 2.0 due to new ways to use OCL_ICD_VENDORS
Vincent Danjean [Thu, 13 Sep 2012 15:13:04 +0000 (17:13 +0200)]
Add documentation, mainly for envvar
Vincent Danjean [Wed, 12 Sep 2012 22:44:09 +0000 (00:44 +0200)]
Add a test for the Intel workaround
We do not really test the bug itself (would require Intel ICD)
But we test that we run ICD without asking for extensions
Vincent Danjean [Wed, 12 Sep 2012 21:04:19 +0000 (23:04 +0200)]
Workaround for a bug in intel ICD with optimus
Vincent Danjean [Wed, 12 Sep 2012 21:01:21 +0000 (23:01 +0200)]
Cleanup and fix debug ifdef
Vincent Danjean [Tue, 11 Sep 2012 12:07:34 +0000 (14:07 +0200)]
Improve OCL_ICD_VENDORS envvar
Vincent Danjean [Mon, 10 Sep 2012 13:29:46 +0000 (15:29 +0200)]
Force order in output of check programs
Vincent Danjean [Mon, 10 Sep 2012 13:00:30 +0000 (15:00 +0200)]
fix 'make distcheck'
* atenv.m4 was not ditributed
* make installcheck was wrong (should be rewriten)
* atconfig must be removed at distclean