Problem: Some tests assume that all local work-items can be used in a
single dimension of an NDRange.
Spec References: OpenCL C 2.0 r19, table 4.3,
CL_DEVICE_MAX_WORK_ITEM_SIZES.
Solution: The overall maximum local work size is trimmed to that of an
NDRange's first dimension or all dimensions, as appropriate.
Test Suite Affected: atomics, non_uniform_work_group, and workgroups.
Side Effects: None
Change-Id: I2e8179ca15c2c090f47ea84d1d3c109dd69ec185
Until now, the workgroups test computed the work-group size for
execution by querying the maximum number of work-items based on the
kernel. This returns the overall limit on workgroup size but not the
limits in each specific dimension. This change additionally restricts
the work-group size to the device's dimensional limits.
Problem: The kernel's maximum work-group size is used to execute it in a
single dimension, without regard for dimensional limitations.
Spec References: OpenCL 2.0 rev 19
Solution: The kernel's maximum work-group size is decreased such that it
meets any dimensional limitations before it's enqueued.
Test Suite Affected: Workgroups
Side Effects: None
Change-Id: I1c642af087f1934723ac7218ca9a35413ff83224
This change refactors the setting of the compilation mode, so that
instead of using a 'gOfflineCompiler' bool together with a
'gOfflineCompilerOutputType' enum, a single 'gCompilationMode' enum
is used. The default value for gCompilationMode is 'kOnline', which
is equivalent to the previous defaulting of gOfflineCompiler to false.
In addition, it refactors the setting of the compilation cache mode,
so that instead of the 'gForceSpirVCache' and 'gForceSpirVGenerate'
bools, a single 'gCompilationCacheMode' enum is used. The default
value for gCompilationCacheMode is 'kCacheModeCompileIfAbsent', which
is equivalent to the previous defaulting of both booleans to false.
This change also refactors create_openclcpp_program() to avoid saving
and restoring gOfflineCompiler and gOfflineCompilerOutputType. Instead,
it now passes the desired compilation mode as a parameter.
gSpirVPath controls the location of the compilation cache for SPIR-V
offline compilation, but its name falsely implies that it is
SPIR-V specific. This change renames it to gCompilationCachePath.
This change refactors the setting of the compilation mode, so that
instead of using a 'gOfflineCompiler' bool together with a
'gOfflineCompilerOutputType' enum, a single 'gCompilationMode' enum
is used. The default value for gCompilationMode is 'kOnline', which
is equivalent to the previous defaulting of gOfflineCompiler to false.
In addition, it refactors the setting of the compilation cache mode,
so that instead of the 'gForceSpirVCache' and 'gForceSpirVGenerate'
bools, a single 'gCompilationCacheMode' enum is used. The default
value for gCompilationCacheMode is 'kCacheModeCompileIfAbsent', which
is equivalent to the previous defaulting of both booleans to false.
This change removes support for the "source" mode of offline compilation
which is commented with "to be removed in the future as we will use
offline compiler here".
The "source" mode allowed for the CL source to be transformed by
build_script_source.py and then passed to clCreateProgramWithSource(),
and appears to have been developed only for the purpose of testing the
offline compilation infrastructure.
This change removes support for the "source" mode of offline compilation
which is commented with "to be removed in the future as we will use
offline compiler here".
The "source" mode allowed for the CL source to be transformed by
build_script_source.py and then passed to clCreateProgramWithSource(),
and appears to have been developed only for the purpose of testing the
offline compilation infrastructure.
Currently, test_host_queue_order relies on bit accurate float ops.
It expects expects that sqrt((float)i * i) == i.
This is not always true due to floating point precision limitations.
The test does not really need the sqrt operation and it can be removed
instead of trying to correct the floating point check.
Replace problematic float operation with constant value of 1.
Currently, test_host_queue_order relies on bit accurate float ops.
It expects expects that sqrt((float)i * i) == i.
This is not always true due to floating point precision limitations.
The test does not really need the sqrt operation and it can be removed
instead of trying to correct the floating point check.
Replace problematic float operation with constant value of 1.
This adds functionality to define minimal required version through the
ADD_TEST* macros. Tests that don't meet the version requirement will
be skipped.
By default the minimal required version is set to 1.0, subsequent
patches will set the appropriate version for each of the tests.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
This adds functionality to define minimal required version through the
ADD_TEST* macros. Tests that don't meet the version requirement will
be skipped.
By default the minimal required version is set to 1.0, subsequent
patches will set the appropriate version for each of the tests.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
* Fix Issue 38 - Test spirv_new - new test for cl_khr_spirv_no_integer_wrap_decoration SPV_KHR_no_integer_wrap_decoration extension
* Remove not used variable.
This moves from numeric values to enum values to keep result state of
the test, so we can easily save this information in JSON results.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
This moves from numeric values to enum values to keep result state of
the test, so we can easily save this information in JSON results.
Signed-off-by: Radek Szymanski <radek.szymanski@arm.com>
The OpenCL SPIR-V Environment Specification v2.2-10, Section 2.1
states that:
For all OpTypeImage type-declaration instructions:
The optional image Access Qualifier must be present.
Add the ReadOnly or WriteOnly qualifier as appropriate to the .spvasm
files, and regenerate the .spv files using
spirv-as --target-env spv1.0
The input buffer to tests in execute_block suite is set to some
value (0xdeadbeef) but supposed to be reset in each test such
so that observing 0 after executing the test guarantees that
the status is that of the test and not a pre-existing value.
This commit adds missing initialisation to block_barrier kernel code
and removes an extra redundant barrier at the end.
This test is the last in the suite. After executing other tests the
input/output buffer is not being reset on a host side and therefore
running the test after successful execution of the previous test
(resulting in writing expected 0 into the buffer) was hiding the
problem.
Change-Id: Ia9310d095993703c32ee1f664aed9ff72b26367a
Signed-off-by: Anastasia Stulova <anastasia.stulova@arm.com>
Signed-off-by: Kevin Petit <kevin.petit@arm.com>
