Parallel printf tests fail because they use the same file /tmp/tmpfile
to capture OpenCL kernel outputs. This patch solves this problem by
using a unique file for each process.
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Spir test compares floating-point kernel results bit-by-bit with
correct results. However, for math_brute_force kernels, specification
does not ask for SPIR and OpenCL C path to match bit-to-bit. This patch
reuses ulps threshold from math_brute_force folder for math_brute_force
kernels in spir test.
Signed-off-by: Wenju He <wenju.he@intel.com>
* cl12: Compiler - added missing extensions to compiler_defines_for_extensions.
Updated the list of known extensions to include all extensions referenced in the OpenCL Extension Specification 1.2 - Revision 25 and OpenCL Specification 1.2 - Revision 19 documents.
* Re-added the cl_khr_il_program extension.
This was this was erroneously removed in the previous commit.
* Compiler: Added all known extensions to compiler_defines_for_extensions.
After discussion on the pull request, it was decided to have this test test for all known khr extensions accross all API levels. Some vendors support 2.x extensions on 1.x implementations, so wanted to ensure that the test still covered those cases.
Re-ordered each sub-list so that the order in which each extension appears mirrors that of https://www.khronos.org/registry/OpenCL/specs/2.2/html/OpenCL_Ext.html.
For reference: https://github.com/KhronosGroup/OpenCL-CTS/pull/55.
* Compiler: Moved priority and throttle hints to the API-only section of the array in compiler_defines_for_extensions.
* Compiler: Added cl_khr_spirv_no_integer_wrap_decoration to the list of extensions.
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Some of the setup functionality is already there in the test harness, so
use that and remove the duplicated code from within the suite.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
Function verifyOutputBuffer returns non-zero value in case of mismatch.
Return value 'false' means some mismatches are treated as matches. This
commit fix this problem.
The maintenance of the conformance tests is moving to Github.
This commit contains all the changes that have been done in
Gitlab since the first public release of the conformance tests.
Signed-off-by: Kevin Petit <kevin.petit@arm.com>
Performs a single build on Linux using the unified
headers and linking against the ICD.
Unified headers are broken for 2.x on OSX (see
https://github.com/KhronosGroup/OpenCL-Headers/issues/31).
Adding OSX builds should hopefully be a one-line change
once this is fixed.
Signed-off-by: Kevin Petit <kevin.petit@arm.com>
This will always override anything the user provides. We could
turn it into a cache variable to avoid this but the default
value assumes the location of headers in a hypothetical driver
tree which probably doesn't work for anybody so just get rid of
it instead.
Also remove all references to OPENCL_DRIVER in the build system.
Any such integration doesn't really belong in the CTS's build
system and can triavialy be done with a wrapper project.
Signed-off-by: Kevin Petit <kevin.petit@arm.com>
