While #2409 is under review, could we please add "cl_khr_spirv_queries"
to the list of known extensions? This will prevent test "failures" for
implementations that support the extension.
Modify the prime numbers generator function to return `size_t` instead
of `int` and use `0` as a sentinal value instead of hardcoded negative
ones. `0` is not a prime number, so it is suitable to use to indicate an
error.
Fixes#1159
---------
Signed-off-by: Ahmed Hesham <ahmed.hesham@arm.com>
`clGetDeviceInfo` should fail with `CL_INVALID_VALUE` when queried for
`CL_DEVICE_SPIR_VERSIONS` on devices that do not claim to support the
extension that provides it, `cl_khr_spir`.
Following this change, the test is skipped instead of failing on devices
that do not support `cl_khr_spir`.
Signed-off-by: Ahmed Hesham <ahmed.hesham@arm.com>
- Do not override parent-provided value for CMAKE_CXX_FLAGS
- Stop building with -fpermissive (not required/bad practice)
- Delete unused variables
- Remove unnecessary casts
Signed-off-by: Kévin Petit <kpet@free.fr>
fixes#2387
Corrects the "correctly rounded" behavior for the math bruteforce tests.
Specifically:
* Only applies the `-cl-fp32-correctly-rounded-divide-sqrt` build option
for the `divide_cr` and `sqrt_cr` tests. The other tests do not receive
this build option. This means that there is a difference in the behavior
of the `divide` and `divide_cr` tests and the `sqrt` and `sqrt_cr`
tests, and the "correctly rounded" build option is not applied to the
fp16 or fp64 tests.
* Removes the build option to toggle testing the correctly rounded
divide and square root tests since it no longer needed. Instead, the
test names can be used to choose whether to test the correctly rounded
functions or the non-correctly rounded functions.
Additionally:
* Relaxes the fp16 sqrt accuracy requirements to 1 ULP. This is needed
to pass this test on some of our devices. This part is still under
discussion, so I will keep this PR as a draft until it is settled.
Currently, selecting a different device in a platform to test is rather
cumbersome, for two reasons:
1. The default device type tested is the "default" device and there is
at most one default device in a platform. This means that, by itself,
choosing any non-zero device index is by definition out-of-range:
```sh
$ CL_PLATFORM_INDEX=1 CL_DEVICE_INDEX=1 ./test_conformance/basic/test_basic
Initializing random seed to 0.
Requesting Default device based on command line for platform index 1 and device index 1
device index out of range -- choosen_device_index (1) >= num_devices (1)
```
2. To choose a non-default device type you therefore need to explicitly
specify another device type also, but "all" is not a valid device type
in the harness. This means that you need to know both the device type
and the index of the device within that device type to choose the device
to test.
```sh
$ CL_DEVICE_TYPE=all CL_PLATFORM_INDEX=1 CL_DEVICE_INDEX=1 ./test_conformance/basic/test_basic
Unknown CL_DEVICE_TYPE env variable setting: all.
Aborting...
Aborted (core dumped)
```
This PR aims to fix (2), by allowing "all" as a device type. In the
future, we could consider making the default device type "all" vs.
"default", which would fix (1) also, but that will likely need more
discussion and should be done in a separate PR.
* Ulp_Error*: ilogb(reference) - 1 may overflow if reference is zero.
* binary_i_double Test: DoubleFromUInt32's result is a cl_double and the
attempt is to store it as a cl_double, but p was defined as a pointer to
cl_ulong, resulting in an unintended implicit conversion that is not
valid for out-of-range doubles.
* exp2, tanpi: ensure early exit for NaN.
* shift_right_sticky_128: avoid out-of-range shift if shift value is
exactly 64.
* scalbn: e += n may overflow if n is large, move it after the check for
large n.
Add cl_khr_command_buffer test that is it valid to release a
command-buffer after it has been enqueued but before execution is
finished.
This stresses the semantics from
[clReleaseCommandBufferKHR](https://registry.khronos.org/OpenCL/sdk/3.0/docs/man/html/clReleaseCommandBufferKHR.html#_description)
that: "After the command_buffer reference count becomes zero **and has
finished execution**, the command-buffer is deleted"
The `local_workgroup_size` is changed by
`get_max_common_work_group_size`. But the input data still use the
original `local_workgroup_size`, which will cause the check logic to
failed.
Refactor the following tests:
1. `test_imagecopy`
2. `test_imagecopy3d`
3. `test_imagereadwrite`
4. `test_imagereadwrite3d`
The change does the following:
1. Use RAII to manage allocated resources
2. For `imagecopy` and `imagecopy3d`, the change allows for a custom src
image memory flags and adjusts how the source image is created according
to the input flags.
Signed-off-by: Michael Rizkalla <michael.rizkalla@arm.com>
In CMake 3.24+, there is built-in support for adding -Werror that does
not require adding -Werror explicitly, and allows it to be downgraded to
a warning if the user wants that. Use this, to account for warnings that
have false positives.
This change refactors the following tests to use RAII to clean-up
allocated resources on exit, and adds additional changes as mentioned
below:
- test_arrayreadwrite
- Allow different `cl_mem_flags` to be passed to the test.
- test_bufferreadwriterect:
- Allow different `cl_mem_flags` to be passed to the test.
- Customisable copy, read and write functions.
- test_buffer_copy
- Fill the destination buffer with `invalid_ptr` instead of `out_ptr` if
created with `CL_MEM_(USE/COPY)_HOST_PTR`.
- test_buffer_partial_copy
- Fill the destination buffer with `invalid_ptr` instead of `out_ptr` if
created with `CL_MEM_(USE/COPY)_HOST_PTR`.
---------
Signed-off-by: Michael Rizkalla <michael.rizkalla@arm.com>
This commit fixes three tests that have memory leaks due to unreleased
CL objects.
- `kernel_local_memory_size`
- `multi_queue_flush_on_release`
- `queue_flush_on_release`
Signed-off-by: Michael Rizkalla <michael.rizkalla@arm.com>
Found while developing #2376.
Parts of "test_opencl_c_versions" were intending to build programs for
different OpenCL C versions, but unfortunately the helper function the
test was calling only created the program and did not build it. Switch
to a helper function that builds the program, instead.
This change provides partial test coverage for
KhronosGroup/OpenCL-Docs#1280
Adding CTS tests for:
1. clEnqueueMapBuffer, clEnqueueMapImage.
2. Command buffer negative tests.
3. clSetKernelArgs negative tests.
The bulk of the tests is to make sure that the CL driver does not allow
writing to a memory object that is created with `CL_MEM_IMMUTABLE_EXT`
flag when used with the above APIs.
---------
Signed-off-by: Michael Rizkalla <michael.rizkalla@arm.com>
Remove the test timing functionality from the conversions test, because
it
- is not advertised in the help,
- is not supported on all platforms, triggering warnings during the
build,
- goes beyond the core purpose of the conformance test suite.
Signed-off-by: Sven van Haastregt <sven.vanhaastregt@arm.com>
Fixes#2155 according to issue description
Additional remark: The image size was previously calculated based on the
memory size, which seems unusual. Due to Vulkan's configuration
capabilities, the size of memory allocated for a specific texture may
differ from what would be expected based on the texture dimensions.
Thus, calculating the image dimensions back from the memory size of a
Vulkan texture can be challenging.
This change add the following image formats to the supported image
formats:
1. CL_UNSIGNED_INT10X6_EXT
2. CL_UNSIGNED_INT12X4_EXT
3. CL_UNSIGNED_INT14X2_EXT
4. CL_UNORM_INT10X6_EXT
5. CL_UNORM_INT12X4_EXT
6. CL_UNORM_INT14X2_EXT
For reference, the specification for
`cl_ext_image_unsigned_10x6_12x4_14x2` can be found here:
https://github.com/KhronosGroup/OpenCL-Docs/pull/1352
Signed-off-by: Michael Rizkalla <michael.rizkalla@arm.com>
Co-authored-by: Michael Rizkalla <michael.rizkalla@arm.com>
Related to #2142, according to the work plan extended `CBasicTestStore`
with support for `atomic_half`.
Optimization remark: in tests related to `CBasicTestStore` kernel source
code is mostly composed with arguments following similar pattern:
`__kernel void test_atomic_kernel(uint threadCount, uint numDestItems,
__global int *finalDest, __global int *oldValues, volatile __local
atomic_int *destMemory)`
`oldValues` buffer is initialized with a host pointer, after kernel
execution it is read back to the host pointer but it is unused in
neither of the kernels I verified.
- Cases with a NULL global_work_size were not tested
- The specification states that zero-sized enqueues behave similarly to
marker commands. Test that events for NDRange kernel commands have the
correct CL_COMMAND_NDRANGE_KERNEL command type to guard against
implementations naively using a marker command.
- Tidy up the code a bit and add missing error checking.
Signed-off-by: Kévin Petit <kpet@free.fr>
`test_spir` and its data are not installed correctly when building the
installation target.
This change makes sure both the binary and the test data are installed
correctly to `${CMAKE_INSTALL_BINDIR}/$<CONFIG>`.
Signed-off-by: Michael Rizkalla <michael.rizkalla@arm.com>
Call `getAllowedUlpError` to obtain the allowed ULP error for all of the
double type (fp64) tests. The aim is to standardise obtaining the
desired ULP requirement and pave the way for adding the Embedded Profile
ULP errors.
Contributes to https://github.com/KhronosGroup/OpenCL-CTS/issues/867
Signed-off-by: Sven van Haastregt <sven.vanhaastregt@arm.com>
Remove the `CREATE_OPENCL_SEMAPHORE` macro and use derived class
instantiations of the `clExternalSemaphore` class, rather than base
pointers to derived class objects.
Remove the default argument for `queryParamName` in
`check_external_semaphore_handle_type()`.
Move `check_external_semaphore_handle_type()` checks to constructors of
`clExternalImportableSemaphore` and `clExternalExportableSemaphore`,
rather than manually making the check before creating an external
semaphore.
---------
Signed-off-by: Gorazd Sumkovski <gorazd.sumkovski@arm.com>
Co-authored-by: Kévin Petit <kpet@free.fr>
Co-authored-by: Kevin Petit <kevin.petit@arm.com>
Using C++17's `if constexpr` avoids Wformat warnings in this template
function, as it is now made explicit that the various format strings are
only used with appropriate types.
Signed-off-by: Sven van Haastregt <sven.vanhaastregt@arm.com>
Support for optimally-tiled images and tiling inference in general is
implementation-defined. It should not be relied upon in the CTS.
Also build the code base as C++17 for std::optional. Many Khronos
projects are now using C++17 and the transition in the CTS is IMO
overdue.
Signed-off-by: Kevin Petit <kevin.petit@arm.com>
This enables many useful newer C++ features. C++17 is already required
by many components of the Khronos ecosystem.
Signed-off-by: Kevin Petit <kevin.petit@arm.com>
After https://github.com/KhronosGroup/SPIRV-Tools/pull/6027 spirv-val is
now rejecting `Aligned 0` Memory Operands. This causes the spirv_new
test binary to no longer build with a recent SPIRV-Tools version.
Mechanically remove all occurrences of `Aligned 0` in the SPIR-V
assembly files.
Signed-off-by: Sven van Haastregt <sven.vanhaastregt@arm.com>
