🏡 index : cold-air-networks/mirror-opencl-cts.git

📄 .appveyor.yml
📄 .clang-format
📄 .gitattributes
📁 .github
📁 workflows
📄 presubmit.yml
📄 CMakeLists.txt
📄 CODE_OF_CONDUCT.md
📄 LICENSE.txt
📄 README.md
📄 check-format.sh
📄 presubmit.sh
📄 readme-spir-v-binaries.txt
📁 test_common
📄 CMakeLists.txt
📁 autotest
📄 autotest.hpp
📄 test_suite.hpp
📄 config.hpp
📁 gl
📄 gl_headers.h
📄 helpers.cpp
📄 helpers.h
📄 setup.h
📄 setup_osx.cpp
📄 setup_win32.cpp
📄 setup_x11.cpp
📁 gles
📄 gl_headers.h
📄 helpers.cpp
📄 helpers.h
📄 setup.h
📁 harness
📄 ThreadPool.cpp
📄 ThreadPool.h
📄 alloc.h
📄 clImageHelper.h
📄 cl_offline_compiler-interface.txt
📄 compat.h
📄 conversions.cpp
📄 conversions.h
📄 crc32.cpp
📄 crc32.h
📄 deviceInfo.cpp
📄 deviceInfo.h
📄 errorHelpers.cpp
📄 errorHelpers.h
📄 featureHelpers.cpp
📄 featureHelpers.h
📄 fpcontrol.h
📄 genericThread.cpp
📄 genericThread.h
📄 imageHelpers.cpp
📄 imageHelpers.h
📄 kernelHelpers.cpp
📄 kernelHelpers.h
📄 mingw_compat.c
📄 mingw_compat.h
📄 msvc9.c
📄 mt19937.cpp
📄 mt19937.h
📄 os_helpers.cpp
📄 os_helpers.h
📄 parseParameters.cpp
📄 parseParameters.h
📄 propertyHelpers.cpp
📄 propertyHelpers.h
📄 ref_counting.h
📄 rounding_mode.cpp
📄 rounding_mode.h
📄 testHarness.cpp
📄 testHarness.h
📄 test_mt19937.c
📄 threadTesting.cpp
📄 threadTesting.h
📄 typeWrappers.cpp
📄 typeWrappers.h
📁 miniz
📄 CMakeLists.txt
📄 miniz.c
📄 miniz.h
📁 test_conformance
📄 CMakeCommon.txt
📄 CMakeLists.txt
📁 SVM
📄 CMakeLists.txt
📄 common.h
📄 main.cpp
📄 test_allocate_shared_buffer.cpp
📄 test_byte_granularity.cpp
📄 test_cross_buffer_pointers.cpp
📄 test_enqueue_api.cpp
📄 test_fine_grain_memory_consistency.cpp
📄 test_fine_grain_sync_buffers.cpp
📄 test_migrate.cpp
📄 test_pointer_passing.cpp
📄 test_set_kernel_exec_info_svm_ptrs.cpp
📄 test_shared_address_space_coarse_grain.cpp
📄 test_shared_address_space_fine_grain.cpp
📄 test_shared_address_space_fine_grain_buffers.cpp
📄 test_shared_sub_buffers.cpp
📁 allocations
📄 CMakeLists.txt
📄 allocation_execute.cpp
📄 allocation_execute.h
📄 allocation_fill.cpp
📄 allocation_fill.h
📄 allocation_functions.cpp
📄 allocation_functions.h
📄 allocation_utils.cpp
📄 allocation_utils.h
📄 main.cpp
📄 testBase.h
📁 api
📄 CMakeLists.txt
📄 main.cpp
📄 negative_platform.cpp
📄 procs.h
📄 testBase.h
📄 test_api_consistency.cpp
📄 test_api_min_max.cpp
📄 test_binary.cpp
📄 test_bool.cpp
📄 test_clone_kernel.cpp
📄 test_context_destructor_callback.cpp
📄 test_create_context_from_type.cpp
📄 test_create_kernels.cpp
📄 test_device_min_data_type_align_size_alignment.cpp
📄 test_kernel_arg_changes.cpp
📄 test_kernel_arg_info.cpp
📄 test_kernel_arg_info_compatibility.cpp
📄 test_kernel_arg_multi_setup.cpp
📄 test_kernel_private_memory_size.cpp
📄 test_kernels.cpp
📄 test_mem_object_info.cpp
📄 test_mem_object_properties_queries.cpp
📄 test_mem_objects.cpp
📄 test_min_image_formats.cpp
📄 test_native_kernel.cpp
📄 test_null_buffer_arg.cpp
📄 test_pipe_properties_queries.cpp
📄 test_platform.cpp
📄 test_queries.cpp
📄 test_queue.cpp
📄 test_queue_hint.cpp
📄 test_queue_properties.cpp
📄 test_queue_properties_queries.cpp
📄 test_retain.cpp
📄 test_retain_program.cpp
📄 test_sub_group_dispatch.cpp
📄 test_zero_sized_enqueue.cpp
📁 atomics
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 testBase.h
📄 test_atomics.cpp
📄 test_indexed_cases.cpp
📁 basic
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 run_array
📄 run_array_image_copy
📄 run_image
📄 run_multi_read_image
📄 test_arraycopy.cpp
📄 test_arrayimagecopy.cpp
📄 test_arrayimagecopy3d.cpp
📄 test_arrayreadwrite.cpp
📄 test_astype.cpp
📄 test_async_copy.cpp
📄 test_async_copy2D.cpp
📄 test_async_copy3D.cpp
📄 test_async_copy_fence.cpp
📄 test_async_strided_copy.cpp
📄 test_barrier.cpp
📄 test_basic_parameter_types.cpp
📄 test_bufferreadwriterect.cpp
📄 test_constant.cpp
📄 test_constant_source.cpp
📄 test_createkernelsinprogram.cpp
📄 test_enqueue_map.cpp
📄 test_enqueued_local_size.cpp
📄 test_explicit_s2v.cpp
📄 test_float2int.cpp
📄 test_fpmath_float.cpp
📄 test_fpmath_float2.cpp
📄 test_fpmath_float4.cpp
📄 test_get_linear_ids.cpp
📄 test_global_linear_id.cpp
📄 test_global_work_offsets.cpp
📄 test_hiloeo.cpp
📄 test_hostptr.cpp
📄 test_if.cpp
📄 test_imagecopy.cpp
📄 test_imagecopy3d.cpp
📄 test_imagedim.cpp
📄 test_imagenpot.cpp
📄 test_imagerandomcopy.cpp
📄 test_imagereadwrite.cpp
📄 test_imagereadwrite3d.cpp
📄 test_int2float.cpp
📄 test_intmath.cpp
📄 test_kernel_call_kernel_function.cpp
📄 test_kernel_memory_alignment.cpp
📄 test_local.cpp
📄 test_local_kernel_scope.cpp
📄 test_local_linear_id.cpp
📄 test_loop.cpp
📄 test_multireadimagemultifmt.cpp
📄 test_multireadimageonefmt.cpp
📄 test_numeric_constants.cpp
📄 test_pointercast.cpp
📄 test_preprocessors.cpp
📄 test_progvar.cpp
📄 test_queue_priority.cpp
📄 test_readimage.cpp
📄 test_readimage3d.cpp
📄 test_readimage3d_fp32.cpp
📄 test_readimage3d_int16.cpp
📄 test_readimage_fp32.cpp
📄 test_readimage_int16.cpp
📄 test_rw_image_access_qualifier.cpp
📄 test_simple_image_pitch.cpp
📄 test_sizeof.cpp
📄 test_vec_type_hint.cpp
📄 test_vector_creation.cpp
📄 test_vector_swizzle.cpp
📄 test_vloadstore.cpp
📄 test_wg_barrier.cpp
📄 test_work_item_functions.cpp
📄 test_writeimage.cpp
📄 test_writeimage_fp32.cpp
📄 test_writeimage_int16.cpp
📁 buffers
📄 CMakeLists.txt
📄 array_info.cpp
📄 main.cpp
📄 procs.h
📄 test_buffer_copy.cpp
📄 test_buffer_fill.cpp
📄 test_buffer_map.cpp
📄 test_buffer_mem.cpp
📄 test_buffer_migrate.cpp
📄 test_buffer_read.cpp
📄 test_buffer_write.cpp
📄 test_image_migrate.cpp
📄 test_sub_buffers.cpp
📁 c11_atomics
📄 CMakeLists.txt
📄 common.cpp
📄 common.h
📄 host_atomics.cpp
📄 host_atomics.h
📄 main.cpp
📄 test_atomics.cpp
📁 clcpp
📄 CMakeLists.txt
📁 address_spaces
📄 CMakeLists.txt
📄 common.hpp
📄 main.cpp
📄 test_pointer_types.hpp
📄 test_storage_types.hpp
📁 api
📄 CMakeLists.txt
📄 main.cpp
📄 test_ctors.hpp
📄 test_ctors_dtors.hpp
📄 test_dtors.hpp
📄 test_spec_consts.hpp
📁 atomics
📄 CMakeLists.txt
📄 atomic_fetch.hpp
📄 main.cpp
📁 attributes
📄 CMakeLists.txt
📄 main.cpp
📄 test_ivdep.hpp
📄 test_max_size.hpp
📄 test_required_num_sub_groups.hpp
📄 common.hpp
📁 common_funcs
📄 CMakeLists.txt
📄 common_funcs.hpp
📄 main.cpp
📁 convert
📄 CMakeLists.txt
📄 convert_cast.hpp
📄 main.cpp
📁 device_queue
📄 CMakeLists.txt
📄 main.cpp
📄 test_enqueue.hpp
📄 funcs_test_utils.hpp
📁 geometric_funcs
📄 CMakeLists.txt
📄 fast_geometric_funcs.hpp
📄 geometric_funcs.hpp
📄 main.cpp
📁 images
📄 CMakeLists.txt
📄 common.hpp
📄 main.cpp
📄 test_read.hpp
📄 test_sample.hpp
📄 test_write.hpp
📁 integer_funcs
📄 24bit_funcs.hpp
📄 CMakeLists.txt
📄 bitwise_funcs.hpp
📄 common.hpp
📄 main.cpp
📄 numeric_funcs.hpp
📁 math_funcs
📄 CMakeLists.txt
📄 common.hpp
📄 comparison_funcs.hpp
📄 exponential_funcs.hpp
📄 floating_point_funcs.hpp
📄 half_math_funcs.hpp
📄 logarithmic_funcs.hpp
📄 main.cpp
📄 other_funcs.hpp
📄 power_funcs.hpp
📄 reference.hpp
📄 trigonometric_funcs.hpp
📁 pipes
📄 CMakeLists.txt
📄 main.cpp
📄 test_pipes.hpp
📁 program_scope_ctors_dtors
📄 CMakeLists.txt
📄 common.hpp
📄 main.cpp
📄 test_ctors_dtors.hpp
📁 reinterpret
📄 CMakeLists.txt
📄 as_type.hpp
📄 main.cpp
📁 relational_funcs
📄 CMakeLists.txt
📄 common.hpp
📄 comparison_funcs.hpp
📄 main.cpp
📄 select_funcs.hpp
📄 test_funcs.hpp
📁 spec_constants
📄 CMakeLists.txt
📄 common.hpp
📄 main.cpp
📄 test_spec_consts_attributes.hpp
📄 test_spec_consts_if.hpp
📄 test_spec_consts_init_vars.hpp
📄 spirv10_2016.04.27.7z
📁 subgroups
📄 CMakeLists.txt
📄 common.hpp
📄 main.cpp
📄 test_sg_all.hpp
📄 test_sg_any.hpp
📄 test_sg_broadcast.hpp
📄 test_sg_reduce.hpp
📄 test_sg_scan_exclusive.hpp
📄 test_sg_scan_inclusive.hpp
📁 synchronization
📄 CMakeLists.txt
📄 main.cpp
📁 named_barrier
📄 common.hpp
📄 test_named_barrier.hpp
📄 test_spec_example.hpp
📄 test_sub_group_barrier.hpp
📄 test_work_group_barrier.hpp
📁 utils_common
📄 errors.hpp
📄 is_vector_type.hpp
📄 kernel_helpers.hpp
📄 make_vector_type.hpp
📄 scalar_type.hpp
📄 string.hpp
📄 type_name.hpp
📄 type_supported.hpp
📄 vector_size.hpp
📁 utils_test
📄 binary.hpp
📄 compare.hpp
📁 detail
📄 base_func_type.hpp
📄 vec_helpers.hpp
📄 generate_inputs.hpp
📄 ternary.hpp
📄 unary.hpp
📁 vload_vstore
📄 CMakeLists.txt
📄 common.hpp
📄 half_utils.hpp
📄 main.cpp
📄 vload_funcs.hpp
📄 vstore_funcs.hpp
📁 workgroups
📄 CMakeLists.txt
📄 common.hpp
📄 main.cpp
📄 test_wg_all.hpp
📄 test_wg_any.hpp
📄 test_wg_broadcast.hpp
📄 test_wg_reduce.hpp
📄 test_wg_scan_exclusive.hpp
📄 test_wg_scan_inclusive.hpp
📁 workitems
📄 CMakeLists.txt
📄 main.cpp
📄 test_workitems.hpp
📁 commonfns
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 test_binary_fn.cpp
📄 test_clamp.cpp
📄 test_degrees.cpp
📄 test_fmax.cpp
📄 test_fmaxf.cpp
📄 test_fmin.cpp
📄 test_fminf.cpp
📄 test_max.cpp
📄 test_maxf.cpp
📄 test_min.cpp
📄 test_minf.cpp
📄 test_mix.cpp
📄 test_radians.cpp
📄 test_sign.cpp
📄 test_smoothstep.cpp
📄 test_smoothstepf.cpp
📄 test_step.cpp
📄 test_stepf.cpp
📁 compiler
📄 CMakeLists.txt
📁 includeTestDirectory
📄 testIncludeFile.h
📄 main.cpp
📄 procs.h
📁 secondIncludeTestDirectory
📄 testIncludeFile.h
📄 testBase.h
📄 test_async_build.cpp
📄 test_build_helpers.cpp
📄 test_build_options.cpp
📄 test_compile.cpp
📄 test_compiler_defines_for_extensions.cpp
📄 test_feature_macro.cpp
📄 test_image_macro.cpp
📄 test_opencl_c_versions.cpp
📄 test_pragma_unroll.cpp
📄 test_preprocessor.cpp
📄 test_unload_platform_compiler.cpp
📄 test_unload_platform_compiler_resources.hpp
📄 write_kernel.spvasm32
📄 write_kernel.spvasm64
📁 computeinfo
📄 CMakeLists.txt
📄 conforming_version.cpp
📄 device_uuid.cpp
📄 extended_versioning.cpp
📄 main.cpp
📁 contractions
📄 CMakeLists.txt
📄 contractions.cpp
📁 conversions
📄 CMakeLists.txt
📄 Sleep.cpp
📄 Sleep.h
📄 basic_test_conversions.cpp
📄 basic_test_conversions.h
📄 fplib.cpp
📄 fplib.h
📄 run_batch
📄 test_conversions.cpp
📁 d3d10
📄 CMakeLists.txt
📄 buffer.cpp
📄 harness.cpp
📄 harness.h
📄 main.cpp
📄 misc.cpp
📄 texture2d.cpp
📄 texture3d.cpp
📁 d3d11
📄 CMakeLists.txt
📄 buffer.cpp
📄 harness.cpp
📄 harness.h
📄 main.cpp
📄 misc.cpp
📄 texture2d.cpp
📄 texture3d.cpp
📁 device_execution
📄 CMakeLists.txt
📄 device_info.cpp
📄 device_queue.cpp
📄 enqueue_block.cpp
📄 enqueue_flags.cpp
📄 enqueue_multi_queue.cpp
📄 enqueue_ndrange.cpp
📄 enqueue_profiling.cpp
📄 enqueue_wg_size.cpp
📄 execute_block.cpp
📄 host_multi_queue.cpp
📄 host_queue_order.cpp
📄 main.cpp
📄 nested_blocks.cpp
📄 procs.h
📄 utils.cpp
📄 utils.h
📁 device_partition
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 testBase.h
📄 test_device_partition.cpp
📁 device_timer
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 test_device_timer.cpp
📁 events
📄 CMakeLists.txt
📄 action_classes.cpp
📄 action_classes.h
📄 main.cpp
📄 procs.h
📄 testBase.h
📄 test_callbacks.cpp
📄 test_event_dependencies.cpp
📄 test_events.cpp
📄 test_userevents.cpp
📄 test_userevents_multithreaded.cpp
📄 test_waitlists.cpp
📁 extensions
📄 CMakeLists.txt
📁 cl_ext_cxx_for_opencl
📄 CMakeLists.txt
📄 cxx_for_opencl_ext.cpp
📄 cxx_for_opencl_ver.cpp
📄 main.cpp
📄 procs.h
📁 cl_khr_dx9_media_sharing
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 test_create_context.cpp
📄 test_functions_api.cpp
📄 test_functions_kernel.cpp
📄 test_get_device_ids.cpp
📄 test_interop_sync.cpp
📄 test_memory_access.cpp
📄 test_other_data_types.cpp
📄 utils.cpp
📄 utils.h
📄 wrappers.cpp
📄 wrappers.h
📄 generate_spirv_offline.py
📁 generic_address_space
📄 CMakeLists.txt
📄 advanced_tests.cpp
📄 base.h
📄 basic_tests.cpp
📄 main.cpp
📄 stress_tests.cpp
📁 geometrics
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 testBase.h
📄 test_geometrics.cpp
📄 test_geometrics_double.cpp
📁 gl
📄 CMakeLists.txt
📁 GLEW
📁 GL
📄 eglew.h
📄 glew.h
📄 glxew.h
📄 wglew.h
📄 glew.c
📄 common.h
📄 helpers.cpp
📄 main.cpp
📄 procs.h
📄 testBase.h
📄 test_buffers.cpp
📄 test_fence_sync.cpp
📄 test_image_methods.cpp
📄 test_images_1D.cpp
📄 test_images_1Darray.cpp
📄 test_images_2D.cpp
📄 test_images_2Darray.cpp
📄 test_images_3D.cpp
📄 test_images_depth.cpp
📄 test_images_getinfo_common.cpp
📄 test_images_multisample.cpp
📄 test_images_read_common.cpp
📄 test_images_write_common.cpp
📄 test_renderbuffer.cpp
📄 test_renderbuffer_info.cpp
📁 gles
📄 CMakeLists.txt
📄 helpers.cpp
📄 main.cpp
📄 procs.h
📄 setup_egl.cpp
📄 testBase.h
📄 test_buffers.cpp
📄 test_fence_sync.cpp
📄 test_images_2D.cpp
📄 test_images_2D_info.cpp
📄 test_images_3D.cpp
📄 test_images_3D_info.cpp
📄 test_renderbuffer.cpp
📄 test_renderbuffer_info.cpp
📁 half
📄 CMakeLists.txt
📄 Test_roundTrip.cpp
📄 Test_vLoadHalf.cpp
📄 Test_vStoreHalf.cpp
📄 cl_utils.cpp
📄 cl_utils.h
📄 main.cpp
📄 test_config.h
📄 tests.h
📁 images
📄 CMakeLists.txt
📁 clCopyImage
📄 CMakeLists.txt
📄 main.cpp
📄 test_copy_1D.cpp
📄 test_copy_1D_array.cpp
📄 test_copy_2D.cpp
📄 test_copy_2D_2D_array.cpp
📄 test_copy_2D_3D.cpp
📄 test_copy_2D_array.cpp
📄 test_copy_3D.cpp
📄 test_copy_3D_2D_array.cpp
📄 test_copy_generic.cpp
📄 test_loops.cpp
📁 clFillImage
📄 CMakeLists.txt
📄 main.cpp
📄 test_fill_1D.cpp
📄 test_fill_1D_array.cpp
📄 test_fill_2D.cpp
📄 test_fill_2D_array.cpp
📄 test_fill_3D.cpp
📄 test_fill_generic.cpp
📄 test_loops.cpp
📁 clGetInfo
📄 CMakeLists.txt
📄 main.cpp
📄 test_1D.cpp
📄 test_1D_2D_array.cpp
📄 test_2D.cpp
📄 test_3D.cpp
📄 test_loops.cpp
📁 clReadWriteImage
📄 CMakeLists.txt
📄 main.cpp
📄 test_loops.cpp
📄 test_read_1D.cpp
📄 test_read_1D_array.cpp
📄 test_read_2D.cpp
📄 test_read_2D_array.cpp
📄 test_read_3D.cpp
📄 common.cpp
📄 common.h
📁 kernel_image_methods
📄 CMakeLists.txt
📄 main.cpp
📄 test_1D.cpp
📄 test_1D_array.cpp
📄 test_2D.cpp
📄 test_2D_array.cpp
📄 test_3D.cpp
📄 test_loops.cpp
📁 kernel_read_write
📄 CMakeLists.txt
📄 main.cpp
📄 test_common.cpp
📄 test_common.h
📄 test_iterations.cpp
📄 test_loops.cpp
📄 test_read_1D.cpp
📄 test_read_1D_array.cpp
📄 test_read_2D_array.cpp
📄 test_read_3D.cpp
📄 test_write_1D.cpp
📄 test_write_1D_array.cpp
📄 test_write_2D_array.cpp
📄 test_write_3D.cpp
📄 test_write_image.cpp
📁 samplerlessReads
📄 CMakeLists.txt
📄 main.cpp
📄 test_iterations.cpp
📄 test_loops.cpp
📄 test_read_1D.cpp
📄 test_read_1D_array.cpp
📄 test_read_1D_buffer.cpp
📄 test_read_2D_array.cpp
📄 test_read_3D.cpp
📄 testBase.h
📁 integer_ops
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 testBase.h
📄 test_abs.cpp
📄 test_absdiff.cpp
📄 test_add_sat.cpp
📄 test_int_basic_ops.cpp
📄 test_integers.cpp
📄 test_intmad24.cpp
📄 test_intmul24.cpp
📄 test_popcount.cpp
📄 test_sub_sat.cpp
📄 test_unary_ops.cpp
📄 test_upsample.cpp
📄 verification_and_generation_functions.cpp
📁 math_brute_force
📄 CMakeLists.txt
📄 README.txt
📄 binary_double.cpp
📄 binary_float.cpp
📄 binary_i_double.cpp
📄 binary_i_float.cpp
📄 binary_operator_double.cpp
📄 binary_operator_float.cpp
📄 binary_two_results_i_double.cpp
📄 binary_two_results_i_float.cpp
📄 function_list.cpp
📄 function_list.h
📄 i_unary_double.cpp
📄 i_unary_float.cpp
📄 macro_binary_double.cpp
📄 macro_binary_float.cpp
📄 macro_unary_double.cpp
📄 macro_unary_float.cpp
📄 mad_double.cpp
📄 mad_float.cpp
📄 main.cpp
📄 reference_math.cpp
📄 reference_math.h
📄 run_math_brute_force_in_parallel.py
📄 sleep.cpp
📄 sleep.h
📄 ternary_double.cpp
📄 ternary_float.cpp
📄 test_functions.h
📄 unary_double.cpp
📄 unary_float.cpp
📄 unary_two_results_double.cpp
📄 unary_two_results_float.cpp
📄 unary_two_results_i_double.cpp
📄 unary_two_results_i_float.cpp
📄 unary_u_double.cpp
📄 unary_u_float.cpp
📄 utility.cpp
📄 utility.h
📁 mem_host_flags
📄 CMakeLists.txt
📄 C_host_memory_block.h
📄 checker.h
📄 checker_image_mem_host_no_access.hpp
📄 checker_image_mem_host_read_only.hpp
📄 checker_image_mem_host_write_only.hpp
📄 checker_mem_host_no_access.hpp
📄 checker_mem_host_read_only.hpp
📄 checker_mem_host_write_only.hpp
📄 main.cpp
📄 mem_host_buffer.cpp
📄 mem_host_image.cpp
📄 procs.h
📄 testBase.h
📁 multiple_device_context
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 testBase.h
📄 test_multiple_contexts.cpp
📄 test_multiple_devices.cpp
📁 non_uniform_work_group
📄 CMakeLists.txt
📄 TestNonUniformWorkGroup.cpp
📄 TestNonUniformWorkGroup.h
📄 main.cpp
📄 procs.h
📄 test_advanced_2d.cpp
📄 test_advanced_3d.cpp
📄 test_advanced_other.cpp
📄 test_basic.cpp
📄 tools.cpp
📄 tools.h
📄 opencl_conformance_tests_conversions.csv
📄 opencl_conformance_tests_d3d.csv
📄 opencl_conformance_tests_full.csv
📄 opencl_conformance_tests_full_binary.csv
📄 opencl_conformance_tests_full_no_math_or_conversions.csv
📄 opencl_conformance_tests_full_spirv.csv
📄 opencl_conformance_tests_math.csv
📄 opencl_conformance_tests_quick.csv
📁 pipes
📄 CMakeLists.txt
📄 kernels.h
📄 main.cpp
📄 procs.h
📄 test_pipe_info.cpp
📄 test_pipe_limits.cpp
📄 test_pipe_query_functions.cpp
📄 test_pipe_read_write.cpp
📄 test_pipe_readwrite_errors.cpp
📄 test_pipe_subgroups.cpp
📁 printf
📄 CMakeLists.txt
📄 test_printf.cpp
📄 test_printf.h
📄 util_printf.cpp
📁 profiling
📄 CMakeLists.txt
📄 copy.cpp
📄 execute.cpp
📄 execute_multipass.cpp
📄 main.cpp
📄 procs.h
📄 readArray.cpp
📄 readImage.cpp
📄 writeArray.cpp
📄 writeImage.cpp
📁 relationals
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 testBase.h
📄 test_comparisons_double.cpp
📄 test_comparisons_float.cpp
📄 test_relationals.cpp
📄 test_shuffles.cpp
📄 run_conformance.py
📁 select
📄 CMakeLists.txt
📄 test_select.cpp
📄 test_select.h
📄 util_select.cpp
📁 spir
📄 CMakeLists.txt
📄 api.zip
📄 atomics.zip
📄 basic.zip
📄 binary_type.zip
📄 commonfns.zip
📄 compile_and_link.zip
📄 conversions.zip
📄 datagen.cpp
📄 datagen.h
📄 enum_values.zip
📄 exceptions.h
📄 geometrics.zip
📄 half.zip
📄 images_kernel_read_write.zip
📄 images_samplerlessRead.zip
📄 integer_ops.zip
📄 kernel_attributes.zip
📄 kernel_image_methods.zip
📄 kernelargs.cpp
📄 kernelargs.h
📄 khr.csv
📄 main.cpp
📄 math_brute_force.zip
📄 printf.zip
📄 profiling.zip
📄 relationals.zip
📄 run_build_test.cpp
📄 run_build_test.h
📄 run_services.cpp
📄 run_services.h
📄 sampler_enumeration.zip
📄 select.zip
📄 typeinfo.h
📄 vec_align.zip
📄 vec_step.zip
📁 spirv_new
📄 CMakeLists.txt
📄 README.txt
📄 assemble_spirv.py
📄 main.cpp
📄 procs.h
📁 spirv_asm
📄 atomic_dec_global.spvasm32
📄 atomic_dec_global.spvasm64
📄 atomic_inc_global.spvasm32
📄 atomic_inc_global.spvasm64
📄 branch_conditional.spvasm32
📄 branch_conditional.spvasm64
📄 branch_conditional_weighted.spvasm32
📄 branch_conditional_weighted.spvasm64
📄 branch_simple.spvasm32
📄 branch_simple.spvasm64
📄 composite_construct_int4.spvasm32
📄 composite_construct_int4.spvasm64
📄 composite_construct_struct.spvasm32
📄 composite_construct_struct.spvasm64
📄 constant_char_simple.spvasm32
📄 constant_char_simple.spvasm64
📄 constant_double_simple.spvasm32
📄 constant_double_simple.spvasm64
📄 constant_false_simple.spvasm32
📄 constant_false_simple.spvasm64
📄 constant_float_simple.spvasm32
📄 constant_float_simple.spvasm64
📄 constant_half_simple.spvasm32
📄 constant_half_simple.spvasm64
📄 constant_int3_simple.spvasm32
📄 constant_int3_simple.spvasm64
📄 constant_int4_simple.spvasm32
📄 constant_int4_simple.spvasm64
📄 constant_int_simple.spvasm32
📄 constant_int_simple.spvasm64
📄 constant_long_simple.spvasm32
📄 constant_long_simple.spvasm64
📄 constant_short_simple.spvasm32
📄 constant_short_simple.spvasm64
📄 constant_struct_int_char_simple.spvasm32
📄 constant_struct_int_char_simple.spvasm64
📄 constant_true_simple.spvasm64
📄 constant_uchar_simple.spvasm32
📄 constant_uchar_simple.spvasm64
📄 constant_uint_simple.spvasm32
📄 constant_uint_simple.spvasm64
📄 constant_ulong_simple.spvasm32
📄 constant_ulong_simple.spvasm64
📄 constant_ushort_simple.spvasm32
📄 constant_ushort_simple.spvasm64
📄 copy_char_simple.spvasm32
📄 copy_char_simple.spvasm64
📄 copy_double_simple.spvasm32
📄 copy_double_simple.spvasm64
📄 copy_float_simple.spvasm32
📄 copy_float_simple.spvasm64
📄 copy_half_simple.spvasm32
📄 copy_half_simple.spvasm64
📄 copy_int3_simple.spvasm32
📄 copy_int3_simple.spvasm64
📄 copy_int4_simple.spvasm32
📄 copy_int4_simple.spvasm64
📄 copy_int_simple.spvasm32
📄 copy_int_simple.spvasm64
📄 copy_long_simple.spvasm32
📄 copy_long_simple.spvasm64
📄 copy_short_simple.spvasm32
📄 copy_short_simple.spvasm64
📄 copy_struct_int_char_simple.spvasm32
📄 copy_struct_int_char_simple.spvasm64
📄 copy_struct_int_float_simple.spvasm32
📄 copy_struct_int_float_simple.spvasm64
📄 copy_uint_simple.spvasm64
📄 copy_ulong_simple.spvasm32
📄 copy_ulong_simple.spvasm64
📄 copy_ushort_simple.spvasm32
📄 copy_ushort_simple.spvasm64
📄 decorate_aliased.spvasm32
📄 decorate_aliased.spvasm64
📄 decorate_alignment.spvasm32
📄 decorate_alignment.spvasm64
📄 decorate_constant.spvasm32
📄 decorate_constant.spvasm64
📄 decorate_constant_fail.spvasm32
📄 decorate_constant_fail.spvasm64
📄 decorate_cpacked.spvasm32
📄 decorate_cpacked.spvasm64
📄 decorate_restrict.spvasm32
📄 decorate_restrict.spvasm64
📄 decorate_rounding_rte_double_long.spvasm32
📄 decorate_rounding_rte_double_long.spvasm64
📄 decorate_rounding_rte_float_int.spvasm32
📄 decorate_rounding_rte_float_int.spvasm64
📄 decorate_rounding_rtn_double_long.spvasm32
📄 decorate_rounding_rtn_double_long.spvasm64
📄 decorate_rounding_rtn_float_int.spvasm32
📄 decorate_rounding_rtn_float_int.spvasm64
📄 decorate_rounding_rtp_double_long.spvasm32
📄 decorate_rounding_rtp_double_long.spvasm64
📄 decorate_rounding_rtp_float_int.spvasm32
📄 decorate_rounding_rtp_float_int.spvasm64
📄 decorate_rounding_rtz_double_long.spvasm32
📄 decorate_rounding_rtz_double_long.spvasm64
📄 decorate_saturated_conversion_int.spvasm64
📄 decorate_saturated_conversion_short.spvasm32
📄 decorate_saturated_conversion_short.spvasm64
📄 decorate_saturated_conversion_uchar.spvasm32
📄 decorate_saturated_conversion_uchar.spvasm64
📄 decorate_saturated_conversion_uint.spvasm32
📄 decorate_saturated_conversion_uint.spvasm64
📄 decorate_saturated_conversion_ushort.spvasm32
📄 decorate_saturated_conversion_ushort.spvasm64
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fadd_int.spvasm32
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fadd_int.spvasm64
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fadd_uint.spvasm32
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fadd_uint.spvasm64
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fmul_int.spvasm32
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fmul_int.spvasm64
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fmul_uint.spvasm32
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fmul_uint.spvasm64
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fnegate_int.spvasm32
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fnegate_int.spvasm64
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fshiftleft_int.spvasm32
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fshiftleft_int.spvasm64
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fshiftleft_uint.spvasm32
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fshiftleft_uint.spvasm64
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fsub_int.spvasm32
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fsub_int.spvasm64
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fsub_uint.spvasm32
📄 ext_cl_khr_spirv_no_integer_wrap_decoration_fsub_uint.spvasm64
📄 fadd_double.spvasm32
📄 fadd_double.spvasm64
📄 fadd_double2.spvasm32
📄 fadd_double2.spvasm64
📄 fadd_half.spvasm64
📄 fdiv_double.spvasm32
📄 fdiv_double.spvasm64
📄 fdiv_double2.spvasm32
📄 fdiv_double2.spvasm64
📄 fdiv_float.spvasm32
📄 fdiv_float.spvasm64
📄 fdiv_float4.spvasm32
📄 fdiv_float4.spvasm64
📄 fdiv_half.spvasm32
📄 fdiv_half.spvasm64
📄 fmod_double.spvasm32
📄 fmod_double.spvasm64
📄 fmod_double2.spvasm32
📄 fmod_double2.spvasm64
📄 fmod_float.spvasm32
📄 fmod_float.spvasm64
📄 fmod_float4.spvasm32
📄 fmod_float4.spvasm64
📄 fmod_half.spvasm32
📄 fmod_half.spvasm64
📄 fmul_double.spvasm32
📄 fmul_double.spvasm64
📄 fmul_double2.spvasm32
📄 fmul_double2.spvasm64
📄 fmul_float.spvasm32
📄 frem_double.spvasm32
📄 frem_double.spvasm64
📄 frem_double2.spvasm32
📄 frem_double2.spvasm64
📄 frem_float.spvasm32
📄 frem_float.spvasm64
📄 frem_float4.spvasm32
📄 frem_float4.spvasm64
📄 frem_half.spvasm32
📄 frem_half.spvasm64
📄 fsub_double.spvasm32
📄 fsub_double.spvasm64
📄 fsub_double2.spvasm32
📄 fsub_double2.spvasm64
📄 fsub_float.spvasm32
📄 fsub_float.spvasm64
📄 fsub_float4.spvasm32
📄 fsub_float4.spvasm64
📄 fsub_half.spvasm32
📄 fsub_half.spvasm64
📄 label_simple.spvasm32
📄 label_simple.spvasm64
📄 lifetime_simple.spvasm32
📄 lifetime_simple.spvasm64
📄 linkage_export.spvasm32
📄 linkage_export.spvasm64
📄 linkage_import.spvasm32
📄 linkage_import.spvasm64
📄 loop_merge_branch_conditional_dont_unroll.spvasm32
📄 loop_merge_branch_conditional_dont_unroll.spvasm64
📄 loop_merge_branch_conditional_none.spvasm32
📄 loop_merge_branch_none.spvasm32
📄 loop_merge_branch_none.spvasm64
📄 loop_merge_branch_unroll.spvasm32
📄 loop_merge_branch_unroll.spvasm64
📄 op_function_const.spvasm32
📄 op_function_const.spvasm64
📄 op_function_inline.spvasm32
📄 op_function_inline.spvasm64
📄 op_function_noinline.spvasm32
📄 op_function_noinline.spvasm64
📄 op_function_none.spvasm32
📄 op_function_none.spvasm64
📄 op_function_pure.spvasm32
📄 op_function_pure.spvasm64
📄 op_function_pure_ptr.spvasm32
📄 op_function_pure_ptr.spvasm64
📄 op_neg_double.spvasm32
📄 op_neg_double.spvasm64
📄 op_neg_float.spvasm32
📄 op_neg_float.spvasm64
📄 op_neg_float4.spvasm32
📄 op_neg_float4.spvasm64
📄 op_neg_int.spvasm32
📄 op_neg_int.spvasm64
📄 op_neg_int4.spvasm32
📄 op_neg_int4.spvasm64
📄 op_neg_long.spvasm32
📄 op_neg_long.spvasm64
📄 op_neg_short.spvasm32
📄 op_neg_short.spvasm64
📄 op_not_int.spvasm32
📄 op_not_short.spvasm32
📄 op_not_short.spvasm64
📄 op_spec_constant_double_simple.spvasm32
📄 op_spec_constant_double_simple.spvasm64
📄 op_spec_constant_false_simple.spvasm32
📄 op_spec_constant_false_simple.spvasm64
📄 op_spec_constant_float_simple.spvasm32
📄 op_spec_constant_float_simple.spvasm64
📄 op_spec_constant_half_simple.spvasm32
📄 op_spec_constant_half_simple.spvasm64
📄 op_spec_constant_true_simple.spvasm32
📄 op_spec_constant_true_simple.spvasm64
📄 op_spec_constant_uchar_simple.spvasm32
📄 op_spec_constant_uchar_simple.spvasm64
📄 op_spec_constant_uint_simple.spvasm32
📄 op_spec_constant_uint_simple.spvasm64
📄 op_spec_constant_ulong_simple.spvasm32
📄 op_spec_constant_ulong_simple.spvasm64
📄 op_spec_constant_ushort_simple.spvasm32
📄 op_spec_constant_ushort_simple.spvasm64
📄 opaque.spvasm32
📄 opaque.spvasm64
📄 phi_2.spvasm32
📄 phi_2.spvasm64
📄 phi_3.spvasm32
📄 phi_3.spvasm64
📄 phi_4.spvasm32
📄 phi_4.spvasm64
📄 select_if_dont_flatten.spvasm32
📄 select_if_dont_flatten.spvasm64
📄 select_if_flatten.spvasm32
📄 select_switch_flatten.spvasm32
📄 select_switch_flatten.spvasm64
📄 select_switch_none.spvasm32
📄 select_switch_none.spvasm64
📄 undef_char_simple.spvasm32
📄 undef_char_simple.spvasm64
📄 undef_double_simple.spvasm32
📄 undef_double_simple.spvasm64
📄 undef_false_simple.spvasm32
📄 undef_false_simple.spvasm64
📄 undef_float_simple.spvasm32
📄 undef_float_simple.spvasm64
📄 undef_half_simple.spvasm32
📄 undef_half_simple.spvasm64
📄 undef_int3_simple.spvasm32
📄 undef_int3_simple.spvasm64
📄 undef_int4_simple.spvasm32
📄 undef_int4_simple.spvasm64
📄 undef_int_simple.spvasm32
📄 undef_int_simple.spvasm64
📄 undef_long_simple.spvasm32
📄 undef_long_simple.spvasm64
📄 undef_short_simple.spvasm32
📄 undef_short_simple.spvasm64
📄 undef_struct_int_char_simple.spvasm32
📄 undef_struct_int_char_simple.spvasm64
📄 undef_struct_int_float_simple.spvasm32
📄 undef_struct_int_float_simple.spvasm64
📄 undef_struct_struct_simple.spvasm32
📄 undef_struct_struct_simple.spvasm64
📄 undef_true_simple.spvasm32
📄 undef_ulong_simple.spvasm32
📄 undef_ulong_simple.spvasm64
📄 undef_ushort_simple.spvasm32
📄 undef_ushort_simple.spvasm64
📄 unreachable_simple.spvasm32
📄 unreachable_simple.spvasm64
📄 vector_char16_extract.spvasm32
📄 vector_char16_extract.spvasm64
📄 vector_char16_insert.spvasm32
📄 vector_char16_insert.spvasm64
📄 vector_double2_extract.spvasm32
📄 vector_double2_extract.spvasm64
📄 vector_double2_insert.spvasm32
📄 vector_double2_insert.spvasm64
📄 vector_float4_extract.spvasm32
📄 vector_float4_extract.spvasm64
📄 vector_float4_insert.spvasm32
📄 vector_float4_insert.spvasm64
📄 vector_int4_extract.spvasm32
📄 vector_int4_extract.spvasm64
📄 vector_int4_insert.spvasm32
📄 vector_int4_insert.spvasm64
📄 vector_long2_extract.spvasm32
📄 vector_long2_extract.spvasm64
📄 vector_long2_insert.spvasm32
📄 vector_long2_insert.spvasm64
📄 vector_times_scalar_double.spvasm32
📄 vector_times_scalar_double.spvasm64
📄 vector_times_scalar_float.spvasm32
📄 vector_times_scalar_float.spvasm64
📄 testBase.h
📄 test_cl_khr_spirv_no_integer_wrap_decoration.cpp
📄 test_decorate.cpp
📄 test_get_program_il.cpp
📄 test_linkage.cpp
📄 test_op_atomic.cpp
📄 test_op_branch.cpp
📄 test_op_branch_conditional.cpp
📄 test_op_composite_construct.cpp
📄 test_op_constant.cpp
📄 test_op_copy_object.cpp
📄 test_op_fmath.cpp
📄 test_op_function.cpp
📄 test_op_lifetime.cpp
📄 test_op_loop_merge.cpp
📄 test_op_negate.cpp
📄 test_op_opaque.cpp
📄 test_op_phi.cpp
📄 test_op_selection_merge.cpp
📄 test_op_spec_constant.cpp
📄 test_op_undef.cpp
📄 test_op_vector_extract.cpp
📄 test_op_vector_insert.cpp
📄 test_op_vector_times_scalar.cpp
📄 types.hpp
📁 subgroups
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 subgroup_common_kernels.cpp
📄 subgroup_common_kernels.h
📄 subgroup_common_templates.h
📄 subhelpers.h
📄 test_barrier.cpp
📄 test_ifp.cpp
📄 test_queries.cpp
📄 test_subgroup.cpp
📄 test_subgroup_ballot.cpp
📄 test_subgroup_clustered_reduce.cpp
📄 test_subgroup_extended_types.cpp
📄 test_subgroup_non_uniform_arithmetic.cpp
📄 test_subgroup_non_uniform_vote.cpp
📄 test_subgroup_shuffle.cpp
📄 test_subgroup_shuffle_relative.cpp
📄 test_workitem.cpp
📄 submission_details_template.txt
📁 thread_dimensions
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 test_thread_dimensions.cpp
📁 vectors
📄 CMakeLists.txt
📄 defines.h
📄 globals.cpp
📄 main.cpp
📄 procs.h
📄 structs.cpp
📄 structs.h
📄 testBase.h
📄 test_step.cpp
📄 test_vec_align.cpp
📄 type_replacer.cpp
📄 type_replacer.h
📁 workgroups
📄 CMakeLists.txt
📄 main.cpp
📄 procs.h
📄 testBase.h
📄 test_wg_all.cpp
📄 test_wg_any.cpp
📄 test_wg_broadcast.cpp
📄 test_wg_reduce.cpp
📄 test_wg_reduce_max.cpp
📄 test_wg_reduce_min.cpp
📄 test_wg_scan_exclusive_add.cpp
📄 test_wg_scan_exclusive_max.cpp
📄 test_wg_scan_exclusive_min.cpp
📄 test_wg_scan_inclusive_add.cpp
📄 test_wg_scan_inclusive_max.cpp
📄 test_wg_scan_inclusive_min.cpp 
//
// Copyright (c) 2017 The Khronos Group Inc.
// 
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//    http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
#include "harness/compat.h"
#include "harness/kernelHelpers.h"
#include "harness/testHarness.h"

#include <string.h>
#include "cl_utils.h"
#include "tests.h"

#include <CL/cl_half.h>

typedef struct ComputeReferenceInfoF_
{
    float *x;
    cl_ushort *r;
    f2h f;
    cl_ulong i;
    cl_uint lim;
    cl_uint count;
} ComputeReferenceInfoF;

typedef struct ComputeReferenceInfoD_
{
    double *x;
    cl_ushort *r;
    d2h f;
    cl_ulong i;
    cl_uint lim;
    cl_uint count;
} ComputeReferenceInfoD;

typedef struct CheckResultInfoF_
{
    const float *x;
    const cl_ushort *r;
    const cl_ushort *s;
    f2h f;
    const char *aspace;
    cl_uint lim;
    cl_uint count;
    int vsz;
} CheckResultInfoF;

typedef struct CheckResultInfoD_
{
    const double *x;
    const cl_ushort *r;
    const cl_ushort *s;
    d2h f;
    const char *aspace;
    cl_uint lim;
    cl_uint count;
    int vsz;
} CheckResultInfoD;

static cl_int
ReferenceF(cl_uint jid, cl_uint tid, void *userInfo)
{
    ComputeReferenceInfoF *cri = (ComputeReferenceInfoF *)userInfo;
    cl_uint lim = cri->lim;
    cl_uint count = cri->count;
    cl_uint off = jid * count;
    float *x = cri->x + off;
    cl_ushort *r = cri->r + off;
    f2h f = cri->f;
    cl_ulong i = cri->i + off;
    cl_uint j, rr;

    if (off + count > lim)
        count = lim - off;

    for (j = 0; j < count; ++j) {
        x[j] = as_float((cl_uint)(i + j));
        r[j] = f(x[j]);
    }

    return 0;
}

static cl_int
CheckF(cl_uint jid, cl_uint tid, void *userInfo)
{
    CheckResultInfoF *cri = (CheckResultInfoF *)userInfo;
    cl_uint lim = cri->lim;
    cl_uint count = cri->count;
    cl_uint off = jid * count;
    const float *x = cri->x + off;
    const cl_ushort *r = cri->r + off;
    const cl_ushort *s = cri->s + off;
    f2h f = cri->f;
    cl_uint j;
    cl_ushort correct2 = f( 0.0f);
    cl_ushort correct3 = f(-0.0f);
    cl_int ret = 0;

    if (off + count > lim)
        count = lim - off;

    if (!memcmp(r, s, count*sizeof(cl_ushort)))
        return 0;

    for (j = 0; j < count; j++) {
    if (s[j] == r[j])
        continue;

        // Pass any NaNs
        if ((s[j] & 0x7fff) > 0x7c00 && (r[j] & 0x7fff) > 0x7c00 )
            continue;

        // retry per section 6.5.3.3
        if (IsFloatSubnormal(x[j]) && (s[j] == correct2 || s[j] == correct3))
            continue;

        // if reference result is subnormal, pass any zero
        if (gIsEmbedded && IsHalfSubnormal(r[j]) && (s[j] == 0x0000 || s[j] == 0x8000))
            continue;

        vlog_error("\nFailure at [%u] with %.6a: *0x%04x vs 0x%04x,  vector_size = %d, address_space = %s\n",
                   j+off, x[j], r[j], s[j], cri->vsz, cri->aspace);

        ret = 1;
        break;
    }

    return ret;
}

static cl_int
ReferenceD(cl_uint jid, cl_uint tid, void *userInfo)
{
    ComputeReferenceInfoD *cri = (ComputeReferenceInfoD *)userInfo;
    cl_uint lim = cri->lim;
    cl_uint count = cri->count;
    cl_uint off = jid * count;
    double *x = cri->x + off;
    cl_ushort *r = cri->r + off;
    d2h f = cri->f;
    cl_uint j;
    cl_ulong i = cri->i + off;

    if (off + count > lim)
        count = lim - off;

    for (j = 0; j < count; ++j) {
        x[j] = as_double(DoubleFromUInt((cl_uint)(i + j)));
        r[j] = f(x[j]);
    }

    return 0;
}

static cl_int
CheckD(cl_uint jid, cl_uint tid, void *userInfo)
{
    CheckResultInfoD *cri = (CheckResultInfoD *)userInfo;
    cl_uint lim = cri->lim;
    cl_uint count = cri->count;
    cl_uint off = jid * count;
    const double *x = cri->x + off;
    const cl_ushort *r = cri->r + off;
    const cl_ushort *s = cri->s + off;
    d2h f = cri->f;
    cl_uint j;
    cl_ushort correct2 = f( 0.0);
    cl_ushort correct3 = f(-0.0);
    cl_int ret = 0;

    if (off + count > lim)
        count = lim - off;

    if (!memcmp(r, s, count*sizeof(cl_ushort)))
        return 0;

    for (j = 0; j < count; j++) {
    if (s[j] == r[j])
        continue;

        // Pass any NaNs
        if ((s[j] & 0x7fff) > 0x7c00 && (r[j] & 0x7fff) > 0x7c00)
            continue;

        if (IsDoubleSubnormal(x[j]) && (s[j] == correct2 || s[j] == correct3))
            continue;

        // if reference result is subnormal, pass any zero result
        if (gIsEmbedded && IsHalfSubnormal(r[j]) && (s[j] == 0x0000 || s[j] == 0x8000))
            continue;

        vlog_error("\nFailure at [%u] with %.13la: *0x%04x vs 0x%04x, vector_size = %d, address space = %s (double precision)\n",
                   j+off, x[j], r[j], s[j], cri->vsz, cri->aspace);

        ret = 1;
    break;
    }

    return ret;
}

static cl_half float2half_rte(float f)
{
    return cl_half_from_float(f, CL_HALF_RTE);
}

static cl_half float2half_rtz(float f)
{
    return cl_half_from_float(f, CL_HALF_RTZ);
}

static cl_half float2half_rtp(float f)
{
    return cl_half_from_float(f, CL_HALF_RTP);
}

static cl_half float2half_rtn(float f)
{
    return cl_half_from_float(f, CL_HALF_RTN);
}

static cl_half double2half_rte(double f)
{
    return cl_half_from_double(f, CL_HALF_RTE);
}

static cl_half double2half_rtz(double f)
{
    return cl_half_from_double(f, CL_HALF_RTZ);
}

static cl_half double2half_rtp(double f)
{
    return cl_half_from_double(f, CL_HALF_RTP);
}

static cl_half double2half_rtn(double f)
{
    return cl_half_from_double(f, CL_HALF_RTN);
}

int test_vstore_half( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    switch (get_default_rounding_mode(deviceID))
    {
        case CL_FP_ROUND_TO_ZERO:
            return Test_vStoreHalf_private(deviceID, float2half_rtz, double2half_rte, "");
        case 0:
            return -1;
        default:
            return Test_vStoreHalf_private(deviceID, float2half_rte, double2half_rte, "");
    }
}

int test_vstore_half_rte( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    return Test_vStoreHalf_private(deviceID, float2half_rte, double2half_rte, "_rte");
}

int test_vstore_half_rtz( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    return Test_vStoreHalf_private(deviceID, float2half_rtz, double2half_rtz, "_rtz");
}

int test_vstore_half_rtp( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    return Test_vStoreHalf_private(deviceID, float2half_rtp, double2half_rtp, "_rtp");
}

int test_vstore_half_rtn( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    return Test_vStoreHalf_private(deviceID, float2half_rtn, double2half_rtn, "_rtn");
}

int test_vstorea_half( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    switch (get_default_rounding_mode(deviceID))
    {
        case CL_FP_ROUND_TO_ZERO:
            return Test_vStoreaHalf_private(deviceID,float2half_rtz, double2half_rte, "");
        case 0:
            return -1;
        default:
            return Test_vStoreaHalf_private(deviceID, float2half_rte, double2half_rte, "");
    }
}

int test_vstorea_half_rte( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    return Test_vStoreaHalf_private(deviceID, float2half_rte, double2half_rte, "_rte");
}

int test_vstorea_half_rtz( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    return Test_vStoreaHalf_private(deviceID, float2half_rtz, double2half_rtz, "_rtz");
}

int test_vstorea_half_rtp( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    return Test_vStoreaHalf_private(deviceID, float2half_rtp, double2half_rtp, "_rtp");
}

int test_vstorea_half_rtn( cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements )
{
    return Test_vStoreaHalf_private(deviceID, float2half_rtn, double2half_rtn, "_rtn");
}

#pragma mark -

int Test_vStoreHalf_private( cl_device_id device, f2h referenceFunc, d2h doubleReferenceFunc, const char *roundName )
{
    int vectorSize, error;
    cl_program  programs[kVectorSizeCount+kStrangeVectorSizeCount][3];
    cl_kernel   kernels[kVectorSizeCount+kStrangeVectorSizeCount][3];

    uint64_t time[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
    uint64_t min_time[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
    memset( min_time, -1, sizeof( min_time ) );
    cl_program  doublePrograms[kVectorSizeCount+kStrangeVectorSizeCount][3];
    cl_kernel   doubleKernels[kVectorSizeCount+kStrangeVectorSizeCount][3];
    uint64_t doubleTime[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
    uint64_t min_double_time[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
    memset( min_double_time, -1, sizeof( min_double_time ) );

    bool aligned= false;

    for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
    {
        const char *source[] = {
            "__kernel void test( __global float", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   size_t i = get_global_id(0);\n"
            "   vstore_half",vector_size_name_extensions[vectorSize],roundName,"( p[i], i, f );\n"
            "}\n"
        };

        const char *source_v3[] = {
            "__kernel void test( __global float *p, __global half *f,\n"
            "                   uint extra_last_thread)\n"
            "{\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t last_i = get_global_size(0)-1;\n"
            "   size_t adjust = 0;\n"
            "   if(last_i == i && extra_last_thread != 0) {\n"
            "     adjust = 3-extra_last_thread;\n"
            "   } "
            "   vstore_half3",roundName,"( vload3(i, p-adjust), i, f-adjust );\n"
            "}\n"
        };

        const char *source_private_store[] = {
            "__kernel void test( __global float", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __private ushort data[16];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t offset = 0;\n"
            "   size_t vecsize = vec_step(p[i]);\n"
            "   vstore_half",vector_size_name_extensions[vectorSize],roundName,"( p[i], 0, (__private half *)(&data[0]) );\n"
            "   for(offset = 0; offset < vecsize; offset++)\n"
            "   {\n"
            "       vstore_half(vload_half(offset, (__private half *)data), 0, &f[vecsize*i+offset]);\n"
            "   }\n"
            "}\n"
        };


        const char *source_private_store_v3[] = {
            "__kernel void test( __global float *p, __global half *f,\n"
            "                   uint extra_last_thread )\n"
            "{\n"
            "   __private ushort data[4];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t last_i = get_global_size(0)-1;\n"
            "   size_t adjust = 0;\n"
            "   size_t offset = 0;\n"
            "   if(last_i == i && extra_last_thread != 0) {\n"
            "     adjust = 3-extra_last_thread;\n"
            "   } "
            "   vstore_half3",roundName,"( vload3(i, p-adjust), 0, (__private half *)(&data[0]) );\n"
            "   for(offset = 0; offset < 3; offset++)\n"
            "   {\n"
            "       vstore_half(vload_half(offset, (__private half *) data), 0, &f[3*i+offset-adjust]);\n"
            "   }\n"
            "}\n"
        };

        char local_buf_size[10];
        sprintf(local_buf_size, "%lld", (uint64_t)gWorkGroupSize);


        const char *source_local_store[] = {
            "__kernel void test( __global float", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __local ushort data[16*", local_buf_size, "];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t lid = get_local_id(0);\n"
            "   size_t lsize = get_local_size(0);\n"
            "   size_t vecsize = vec_step(p[0]);\n"
            "   event_t async_event;\n"
            "   vstore_half",vector_size_name_extensions[vectorSize],roundName,"( p[i], lid, (__local half *)(&data[0]) );\n"
            "   barrier( CLK_LOCAL_MEM_FENCE ); \n"
            "   async_event = async_work_group_copy((__global ushort *)f+vecsize*(i-lid), (__local ushort *)(&data[0]), vecsize*lsize, 0);\n" // investigate later
            "   wait_group_events(1, &async_event);\n"
            "}\n"
        };

        const char *source_local_store_v3[] = {
            "__kernel void test( __global float *p, __global half *f,\n"
            "                   uint extra_last_thread )\n"
            "{\n"
            "   __local ushort data[3*(", local_buf_size, "+1)];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t lid = get_local_id(0);\n"
            "   size_t last_i = get_global_size(0)-1;\n"
            "   size_t adjust = 0;\n"
            "   size_t lsize = get_local_size(0);\n"
            "   event_t async_event;\n"
            "   if(last_i == i && extra_last_thread != 0) {\n"
            "     adjust = 3-extra_last_thread;\n"
            "   } "
            "   vstore_half3",roundName,"( vload3(i,p-adjust), lid, (__local half *)(&data[0]) );\n"
            "   barrier( CLK_LOCAL_MEM_FENCE ); \n"
            "   async_event = async_work_group_copy((__global ushort *)(f+3*(i-lid)), (__local ushort *)(&data[adjust]), lsize*3-adjust, 0);\n" // investigate later
            "   wait_group_events(1, &async_event);\n"
            "}\n"
        };

        const char *double_source[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   size_t i = get_global_id(0);\n"
            "   vstore_half",vector_size_name_extensions[vectorSize],roundName,"( p[i], i, f );\n"
            "}\n"
        };

        const char *double_source_private_store[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __private ushort data[16];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t offset = 0;\n"
            "   size_t vecsize = vec_step(p[i]);\n"
            "   vstore_half",vector_size_name_extensions[vectorSize],roundName,"( p[i], 0, (__private half *)(&data[0]) );\n"
            "   for(offset = 0; offset < vecsize; offset++)\n"
            "   {\n"
            "       vstore_half(vload_half(offset, (__private half *)data), 0, &f[vecsize*i+offset]);\n"
            "   }\n"
            "}\n"
        };


        const char *double_source_local_store[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __local ushort data[16*", local_buf_size, "];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t lid = get_local_id(0);\n"
            "   size_t vecsize = vec_step(p[0]);\n"
            "   size_t lsize = get_local_size(0);\n"
            "   event_t async_event;\n"
            "   vstore_half",vector_size_name_extensions[vectorSize],roundName,"( p[i], lid, (__local half *)(&data[0]) );\n"
            "   barrier( CLK_LOCAL_MEM_FENCE ); \n"
            "   async_event = async_work_group_copy((__global ushort *)(f+vecsize*(i-lid)), (__local ushort *)(&data[0]), vecsize*lsize, 0);\n" // investigate later
            "   wait_group_events(1, &async_event);\n"
            "}\n"
        };


        const char *double_source_v3[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double *p, __global half *f ,\n"
            "                   uint extra_last_thread)\n"
            "{\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t last_i = get_global_size(0)-1;\n"
            "   size_t adjust = 0;\n"
            "   if(last_i == i && extra_last_thread != 0) {\n"
            "     adjust = 3-extra_last_thread;\n"
            "   } "
            "   vstore_half3",roundName,"( vload3(i,p-adjust), i, f -adjust);\n"
            "}\n"
        };

        const char *double_source_private_store_v3[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double *p, __global half *f,\n"
            "                   uint extra_last_thread )\n"
            "{\n"
            "   __private ushort data[4];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t last_i = get_global_size(0)-1;\n"
            "   size_t adjust = 0;\n"
            "   size_t offset = 0;\n"
            "   if(last_i == i && extra_last_thread != 0) {\n"
            "     adjust = 3-extra_last_thread;\n"
            "   } "
            "   vstore_half3",roundName,"( vload3(i, p-adjust), 0, (__private half *)(&data[0]) );\n"
            "   for(offset = 0; offset < 3; offset++)\n"
            "   {\n"
            "       vstore_half(vload_half(offset, (__private half *)data), 0, &f[3*i+offset-adjust]);\n"
            "   }\n"
            "}\n"
        };

        const char *double_source_local_store_v3[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double *p, __global half *f,\n"
            "                   uint extra_last_thread )\n"
            "{\n"
            "   __local ushort data[3*(", local_buf_size, "+1)];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t lid = get_local_id(0);\n"
            "   size_t last_i = get_global_size(0)-1;\n"
            "   size_t adjust = 0;\n"
            "   size_t lsize = get_local_size(0);\n"
            "   event_t async_event;\n"
            "   if(last_i == i && extra_last_thread != 0) {\n"
            "     adjust = 3-extra_last_thread;\n"
            "   }\n "
            "   vstore_half3",roundName,"( vload3(i,p-adjust), lid, (__local half *)(&data[0]) );\n"
            "   barrier( CLK_LOCAL_MEM_FENCE ); \n"
            "   async_event = async_work_group_copy((__global ushort *)(f+3*(i-lid)), (__local ushort *)(&data[adjust]), lsize*3-adjust, 0);\n" // investigate later
            "   wait_group_events(1, &async_event);\n"
            "}\n"
        };



        if(g_arrVecSizes[vectorSize] == 3) {
            programs[vectorSize][0] = MakeProgram( device, source_v3, sizeof(source_v3) / sizeof( source_v3[0]) );
        } else {
            programs[vectorSize][0] = MakeProgram( device, source, sizeof(source) / sizeof( source[0]) );
        }
        if( NULL == programs[ vectorSize ][0] )
        {
            gFailCount++;
            return -1;
        }

        kernels[ vectorSize ][0] = clCreateKernel( programs[ vectorSize ][0], "test", &error );
        if( NULL == kernels[vectorSize][0] )
        {
            gFailCount++;
            vlog_error( "\t\tFAILED -- Failed to create kernel. (%d)\n", error );
            return error;
        }

        if(g_arrVecSizes[vectorSize] == 3) {
            programs[vectorSize][1] = MakeProgram( device, source_private_store_v3, sizeof(source_private_store_v3) / sizeof( source_private_store_v3[0]) );
        } else {
            programs[vectorSize][1] = MakeProgram( device, source_private_store, sizeof(source_private_store) / sizeof( source_private_store[0]) );
        }
        if( NULL == programs[ vectorSize ][1] )
        {
            gFailCount++;
            return -1;
        }

        kernels[ vectorSize ][1] = clCreateKernel( programs[ vectorSize ][1], "test", &error );
        if( NULL == kernels[vectorSize][1] )
        {
            gFailCount++;
            vlog_error( "\t\tFAILED -- Failed to create private kernel. (%d)\n", error );
            return error;
        }

        if(g_arrVecSizes[vectorSize] == 3) {
            programs[vectorSize][2] = MakeProgram( device, source_local_store_v3, sizeof(source_local_store_v3) / sizeof( source_local_store_v3[0]) );
            if(  NULL == programs[ vectorSize ][2] )
            {
                unsigned q;
                for ( q= 0; q < sizeof( source_local_store_v3) / sizeof( source_local_store_v3[0]); q++)
                    vlog_error("%s", source_local_store_v3[q]);

                gFailCount++;
                return -1;

            }
        } else {
            programs[vectorSize][2] = MakeProgram( device, source_local_store, sizeof(source_local_store) / sizeof( source_local_store[0]) );
            if( NULL == programs[ vectorSize ][2] )
            {
                unsigned q;
                for ( q= 0; q < sizeof( source_local_store) / sizeof( source_local_store[0]); q++)
                    vlog_error("%s", source_local_store[q]);

                gFailCount++;
                return -1;

            }
        }

        kernels[ vectorSize ][2] = clCreateKernel( programs[ vectorSize ][2], "test", &error );
        if( NULL == kernels[vectorSize][2] )
        {
            gFailCount++;
            vlog_error( "\t\tFAILED -- Failed to create local kernel. (%d)\n", error );
            return error;
        }

        if( gTestDouble )
        {
            if(g_arrVecSizes[vectorSize] == 3) {
                doublePrograms[vectorSize][0] = MakeProgram( device, double_source_v3, sizeof(double_source_v3) / sizeof( double_source_v3[0]) );
            } else {
                doublePrograms[vectorSize][0] = MakeProgram( device, double_source, sizeof(double_source) / sizeof( double_source[0]) );
            }
            if( NULL == doublePrograms[ vectorSize ][0] )
            {
                gFailCount++;
                return -1;
            }

            doubleKernels[ vectorSize ][0] = clCreateKernel( doublePrograms[ vectorSize ][0], "test", &error );
            if( NULL == kernels[vectorSize][0] )
            {
                gFailCount++;
                vlog_error( "\t\tFAILED -- Failed to create double kernel. (%d)\n", error );
                return error;
            }

            if(g_arrVecSizes[vectorSize] == 3)
                doublePrograms[vectorSize][1] = MakeProgram( device, double_source_private_store_v3, sizeof(double_source_private_store_v3) / sizeof( double_source_private_store_v3[0]) );
            else
                doublePrograms[vectorSize][1] = MakeProgram( device, double_source_private_store, sizeof(double_source_private_store) / sizeof( double_source_private_store[0]) );

            if( NULL == doublePrograms[ vectorSize ][1] )
            {
                gFailCount++;
                return -1;
            }

            doubleKernels[ vectorSize ][1] = clCreateKernel( doublePrograms[ vectorSize ][1], "test", &error );
            if( NULL == kernels[vectorSize][1] )
            {
                gFailCount++;
                vlog_error( "\t\tFAILED -- Failed to create double private kernel. (%d)\n", error );
                return error;
            }

            if(g_arrVecSizes[vectorSize] == 3) {
                doublePrograms[vectorSize][2] = MakeProgram( device, double_source_local_store_v3, sizeof(double_source_local_store_v3) / sizeof( double_source_local_store_v3[0]) );
            } else {
                doublePrograms[vectorSize][2] = MakeProgram( device, double_source_local_store, sizeof(double_source_local_store) / sizeof( double_source_local_store[0]) );
            }
            if( NULL == doublePrograms[ vectorSize ][2] )
            {
                gFailCount++;
                return -1;
            }

            doubleKernels[ vectorSize ][2] = clCreateKernel( doublePrograms[ vectorSize ][2], "test", &error );
            if( NULL == kernels[vectorSize][2] )
            {
                gFailCount++;
                vlog_error( "\t\tFAILED -- Failed to create double local kernel. (%d)\n", error );
                return error;
            }
        }
    } // end for vector size

    // Figure out how many elements are in a work block
    size_t elementSize = MAX( sizeof(cl_ushort), sizeof(float));
    size_t blockCount = BUFFER_SIZE / elementSize; // elementSize is power of 2
    uint64_t lastCase = 1ULL << (8*sizeof(float)); // number of floats.
    size_t stride = blockCount;

    if (gWimpyMode)
        stride = (uint64_t)blockCount * (uint64_t)gWimpyReductionFactor;

    // we handle 64-bit types a bit differently.
    if( lastCase == 0 )
        lastCase = 0x100000000ULL;

    uint64_t i, j;
    error = 0;
    uint64_t printMask = (lastCase >> 4) - 1;
    cl_uint count = 0;
    int addressSpace;
    size_t loopCount;
    cl_uint threadCount = GetThreadCount();

    ComputeReferenceInfoF fref;
    fref.x = (float *)gIn_single;
    fref.r = (cl_half *)gOut_half_reference;
    fref.f = referenceFunc;
    fref.lim = blockCount;
    fref.count = (blockCount + threadCount - 1) / threadCount;

    CheckResultInfoF fchk;
    fchk.x = (const float *)gIn_single;
    fchk.r = (const cl_half *)gOut_half_reference;
    fchk.s = (const cl_half *)gOut_half;
    fchk.f = referenceFunc;
    fchk.lim = blockCount;
    fchk.count = (blockCount + threadCount - 1) / threadCount;

    ComputeReferenceInfoD dref;
    dref.x = (double *)gIn_double;
    dref.r = (cl_half *)gOut_half_reference_double;
    dref.f = doubleReferenceFunc;
    dref.lim = blockCount;
    dref.count = (blockCount + threadCount - 1) / threadCount;

    CheckResultInfoD dchk;
    dchk.x = (const double *)gIn_double;
    dchk.r = (const cl_half *)gOut_half_reference_double;
    dchk.s = (const cl_half *)gOut_half;
    dchk.f = doubleReferenceFunc;
    dchk.lim = blockCount;
    dchk.count = (blockCount + threadCount - 1) / threadCount;

    for( i = 0; i < lastCase; i += stride )
    {
        count = (cl_uint) MIN( blockCount, lastCase - i );
        fref.i = i;
        dref.i = i;

        // Compute the input and reference
        ThreadPool_Do(ReferenceF, threadCount, &fref);

        error = clEnqueueWriteBuffer(gQueue, gInBuffer_single, CL_FALSE, 0, count * sizeof(float ), gIn_single, 0, NULL, NULL);
        if (error) {
            vlog_error( "Failure in clWriteBuffer\n" );
            gFailCount++;
            goto exit;
        }

        if (gTestDouble) {
            ThreadPool_Do(ReferenceD, threadCount, &dref);

            error = clEnqueueWriteBuffer(gQueue, gInBuffer_double, CL_FALSE, 0, count * sizeof(double ), gIn_double, 0, NULL, NULL);
            if (error) {
                vlog_error( "Failure in clWriteBuffer\n" );
                gFailCount++;
                goto exit;
            }
        }

        for (vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++) {
            // Loop through vector sizes
            fchk.vsz = g_arrVecSizes[vectorSize];
            dchk.vsz = g_arrVecSizes[vectorSize];

            for ( addressSpace = 0; addressSpace < 3; addressSpace++) {
                // Loop over address spaces
                fchk.aspace = addressSpaceNames[addressSpace];
                dchk.aspace = addressSpaceNames[addressSpace];

                cl_uint pattern = 0xdeaddead;
                memset_pattern4( gOut_half, &pattern, BUFFER_SIZE/2);

                error = clEnqueueWriteBuffer(gQueue, gOutBuffer_half, CL_FALSE,
                                             0, count * sizeof(cl_half),
                                             gOut_half, 0, NULL, NULL);
                if (error) {
                    vlog_error( "Failure in clWriteArray\n" );
                    gFailCount++;
                    goto exit;
                }

                error = RunKernel(device, kernels[vectorSize][addressSpace], gInBuffer_single, gOutBuffer_half,
                                       numVecs(count, vectorSize, aligned) ,
                                  runsOverBy(count, vectorSize, aligned));
                if (error) {
                    gFailCount++;
                    goto exit;
                }

                error = clEnqueueReadBuffer(gQueue, gOutBuffer_half, CL_TRUE, 0,
                                            count * sizeof(cl_half), gOut_half,
                                            0, NULL, NULL);
                if (error) {
                    vlog_error( "Failure in clReadArray\n" );
                    gFailCount++;
                    goto exit;
                }

                error = ThreadPool_Do(CheckF, threadCount, &fchk);
                if (error) {
                            gFailCount++;
                            goto exit;
                        }

                if (gTestDouble) {
                    memset_pattern4( gOut_half, &pattern, BUFFER_SIZE/2);

                    error = clEnqueueWriteBuffer(
                        gQueue, gOutBuffer_half, CL_FALSE, 0,
                        count * sizeof(cl_half), gOut_half, 0, NULL, NULL);
                    if (error) {
                        vlog_error( "Failure in clWriteArray\n" );
                        gFailCount++;
                        goto exit;
                    }

                    error = RunKernel(device, doubleKernels[vectorSize][addressSpace], gInBuffer_double, gOutBuffer_half,
                                      numVecs(count, vectorSize, aligned),
                                      runsOverBy(count, vectorSize, aligned));
                    if (error) {
                        gFailCount++;
                        goto exit;
                    }

                    error = clEnqueueReadBuffer(
                        gQueue, gOutBuffer_half, CL_TRUE, 0,
                        count * sizeof(cl_half), gOut_half, 0, NULL, NULL);
                    if (error) {
                        vlog_error( "Failure in clReadArray\n" );
                        gFailCount++;
                        goto exit;
                    }

                    error = ThreadPool_Do(CheckD, threadCount, &dchk);
                    if (error) {
                                gFailCount++;
                                goto exit;
                            }
                        }
                    }
                }

        if( ((i+blockCount) & ~printMask) == (i+blockCount) )
        {
            vlog( "." );
            fflush( stdout );
        }
    }  // end last case

    loopCount = count == blockCount ? 1 : 100;
    if( gReportTimes )
    {
        //Init the input stream
        cl_float *p = (cl_float *)gIn_single;
        for( j = 0; j < count; j++ )
            p[j] = (float)((double) (rand() - RAND_MAX/2) / (RAND_MAX/2));

        if( (error = clEnqueueWriteBuffer(gQueue, gInBuffer_single, CL_TRUE, 0, count * sizeof( float ), gIn_single, 0, NULL, NULL)) )
        {
            vlog_error( "Failure in clWriteArray\n" );
            gFailCount++;
            goto exit;
        }

        if( gTestDouble )
        {
            //Init the input stream
            cl_double *q = (cl_double *)gIn_double;
            for( j = 0; j < count; j++ )
                q[j] = ((double) (rand() - RAND_MAX/2) / (RAND_MAX/2));

            if( (error = clEnqueueWriteBuffer(gQueue, gInBuffer_double, CL_TRUE, 0, count * sizeof( double ), gIn_double, 0, NULL, NULL)) )
            {
                vlog_error( "Failure in clWriteArray\n" );
                gFailCount++;
                goto exit;
            }
        }

        //Run again for timing
        for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
        {
            uint64_t bestTime = -1ULL;
            for( j = 0; j < loopCount; j++ )
            {
                uint64_t startTime = ReadTime();


                if( (error = RunKernel(device, kernels[vectorSize][0], gInBuffer_single, gOutBuffer_half, numVecs(count, vectorSize, aligned) ,
                                       runsOverBy(count, vectorSize, aligned)) ) )
                {
                    gFailCount++;
                    goto exit;
                }

                if( (error = clFinish(gQueue)) )
                {
                    vlog_error( "Failure in clFinish\n" );
                    gFailCount++;
                    goto exit;
                }
                uint64_t currentTime = ReadTime() - startTime;
                if( currentTime < bestTime )
                    bestTime = currentTime;
                time[ vectorSize ] += currentTime;
            }
            if( bestTime < min_time[ vectorSize ] )
                min_time[ vectorSize ] = bestTime ;

            if( gTestDouble )
            {
                bestTime = -1ULL;
                for( j = 0; j < loopCount; j++ )
                {
                    uint64_t startTime = ReadTime();
                    if( (error = RunKernel(device, doubleKernels[vectorSize][0], gInBuffer_double, gOutBuffer_half, numVecs(count, vectorSize, aligned) ,
                                           runsOverBy(count, vectorSize, aligned)) ) )
                    {
                        gFailCount++;
                        goto exit;
                    }

                    if( (error = clFinish(gQueue)) )
                    {
                        vlog_error( "Failure in clFinish\n" );
                        gFailCount++;
                        goto exit;
                    }
                    uint64_t currentTime = ReadTime() - startTime;
                    if( currentTime < bestTime )
                        bestTime = currentTime;
                    doubleTime[ vectorSize ] += currentTime;
                }
                if( bestTime < min_double_time[ vectorSize ] )
                    min_double_time[ vectorSize ] = bestTime;
            }
        }
    }

    if( gReportTimes )
    {
        for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
            vlog_perf( SubtractTime( time[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) (count * loopCount), 0,
                      "average us/elem", "vStoreHalf%s avg. (%s vector size: %d)", roundName, addressSpaceNames[0], (g_arrVecSizes[vectorSize]) );
        for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
            vlog_perf( SubtractTime( min_time[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) count, 0,
                      "best us/elem", "vStoreHalf%s best (%s vector size: %d)", roundName, addressSpaceNames[0], (g_arrVecSizes[vectorSize])  );
        if( gTestDouble )
        {
            for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
                vlog_perf( SubtractTime( doubleTime[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) (count * loopCount), 0,
                          "average us/elem (double)", "vStoreHalf%s avg. d (%s vector size: %d)", roundName, addressSpaceNames[0],  (g_arrVecSizes[vectorSize])  );
            for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
                vlog_perf( SubtractTime( min_double_time[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) count, 0,
                          "best us/elem (double)", "vStoreHalf%s best d (%s vector size: %d)", roundName, addressSpaceNames[0], (g_arrVecSizes[vectorSize]) );
        }
    }

exit:
    //clean up
    for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
    {
        for ( addressSpace = 0; addressSpace < 3; addressSpace++) {
            clReleaseKernel( kernels[ vectorSize ][ addressSpace ] );
            clReleaseProgram( programs[ vectorSize ][ addressSpace ] );
            if( gTestDouble )
            {
                clReleaseKernel( doubleKernels[ vectorSize ][addressSpace] );
                clReleaseProgram( doublePrograms[ vectorSize ][addressSpace] );
            }
        }
    }

    return error;
}

int Test_vStoreaHalf_private( cl_device_id device, f2h referenceFunc, d2h doubleReferenceFunc, const char *roundName )
{
    int vectorSize, error;
    cl_program  programs[kVectorSizeCount+kStrangeVectorSizeCount][3];
    cl_kernel   kernels[kVectorSizeCount+kStrangeVectorSizeCount][3];

    uint64_t time[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
    uint64_t min_time[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
    memset( min_time, -1, sizeof( min_time ) );
    cl_program  doublePrograms[kVectorSizeCount+kStrangeVectorSizeCount][3];
    cl_kernel   doubleKernels[kVectorSizeCount+kStrangeVectorSizeCount][3];
    uint64_t doubleTime[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
    uint64_t min_double_time[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
    memset( min_double_time, -1, sizeof( min_double_time ) );

    bool aligned = true;

    int minVectorSize = kMinVectorSize;
    // There is no aligned scalar vstorea_half
    if( 0 == minVectorSize )
        minVectorSize = 1;

    //Loop over vector sizes
    for( vectorSize = minVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
    {
        const char *source[] = {
            "__kernel void test( __global float", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   size_t i = get_global_id(0);\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( p[i], i, f );\n"
            "}\n"
        };

        const char *source_v3[] = {
            "__kernel void test( __global float3 *p, __global half *f )\n"
            "{\n"
            "   size_t i = get_global_id(0);\n"
            "   vstorea_half3",roundName,"( p[i], i, f );\n"
            "   vstore_half",roundName,"( ((__global  float *)p)[4*i+3], 4*i+3, f);\n"
            "}\n"
        };

        const char *source_private[] = {
            "__kernel void test( __global float", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __private float", vector_size_name_extensions[vectorSize], " data;\n"
            "   size_t i = get_global_id(0);\n"
            "   data = p[i];\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( data, i, f );\n"
            "}\n"
        };

        const char *source_private_v3[] = {
            "__kernel void test( __global float3 *p, __global half *f )\n"
            "{\n"
            "   __private float", vector_size_name_extensions[vectorSize], " data;\n"
            "   size_t i = get_global_id(0);\n"
            "   data = p[i];\n"
            "   vstorea_half3",roundName,"( data, i, f );\n"
            "   vstore_half",roundName,"( ((__global  float *)p)[4*i+3], 4*i+3, f);\n"
            "}\n"
        };

        char local_buf_size[10];
        sprintf(local_buf_size, "%lld", (uint64_t)gWorkGroupSize);
        const char *source_local[] = {
            "__kernel void test( __global float", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __local float", vector_size_name_extensions[vectorSize], " data[", local_buf_size, "];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t lid = get_local_id(0);\n"
            "   data[lid] = p[i];\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( data[lid], i, f );\n"
            "}\n"
        };

        const char *source_local_v3[] = {
            "__kernel void test( __global float", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __local float", vector_size_name_extensions[vectorSize], " data[", local_buf_size, "];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t lid = get_local_id(0);\n"
            "   data[lid] = p[i];\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( data[lid], i, f );\n"
            "   vstore_half",roundName,"( ((__global float *)p)[4*i+3], 4*i+3, f);\n"
            "}\n"
        };

        const char *double_source[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   size_t i = get_global_id(0);\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( p[i], i, f );\n"
            "}\n"
        };

        const char *double_source_v3[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   size_t i = get_global_id(0);\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( p[i], i, f );\n"
            "   vstore_half",roundName,"( ((__global double *)p)[4*i+3], 4*i+3, f);\n"
            "}\n"
        };

        const char *double_source_private[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __private double", vector_size_name_extensions[vectorSize], " data;\n"
            "   size_t i = get_global_id(0);\n"
            "   data = p[i];\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( data, i, f );\n"
            "}\n"
        };

        const char *double_source_private_v3[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __private double", vector_size_name_extensions[vectorSize], " data;\n"
            "   size_t i = get_global_id(0);\n"
            "   data = p[i];\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( data, i, f );\n"
            "   vstore_half",roundName,"( ((__global  double *)p)[4*i+3], 4*i+3, f);\n"
            "}\n"
        };

        const char *double_source_local[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __local double", vector_size_name_extensions[vectorSize], " data[", local_buf_size, "];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t lid = get_local_id(0);\n"
            "   data[lid] = p[i];\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( data[lid], i, f );\n"
            "}\n"
        };

        const char *double_source_local_v3[] = {
            "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
            "__kernel void test( __global double", vector_size_name_extensions[vectorSize]," *p, __global half *f )\n"
            "{\n"
            "   __local double", vector_size_name_extensions[vectorSize], " data[", local_buf_size, "];\n"
            "   size_t i = get_global_id(0);\n"
            "   size_t lid = get_local_id(0);\n"
            "   data[lid] = p[i];\n"
            "   vstorea_half",vector_size_name_extensions[vectorSize],roundName,"( data[lid], i, f );\n"
            "   vstore_half",roundName,"( ((__global double *)p)[4*i+3], 4*i+3, f);\n"
            "}\n"
        };

        if(g_arrVecSizes[vectorSize] == 3) {
            programs[vectorSize][0] = MakeProgram( device, source_v3, sizeof(source_v3) / sizeof( source_v3[0]) );
            if( NULL == programs[ vectorSize ][0] )
            {
                gFailCount++;
                return -1;
            }
        } else {
            programs[vectorSize][0] = MakeProgram( device, source, sizeof(source) / sizeof( source[0]) );
            if( NULL == programs[ vectorSize ][0] )
            {
                gFailCount++;
                return -1;
            }
        }

        kernels[ vectorSize ][0] = clCreateKernel( programs[ vectorSize ][0], "test", &error );
        if( NULL == kernels[vectorSize][0] )
        {
            gFailCount++;
            vlog_error( "\t\tFAILED -- Failed to create kernel. (%d)\n", error );
            return error;
        }

        if(g_arrVecSizes[vectorSize] == 3) {
            programs[vectorSize][1] = MakeProgram( device, source_private_v3, sizeof(source_private_v3) / sizeof( source_private_v3[0]) );
            if( NULL == programs[ vectorSize ][1] )
            {
                gFailCount++;
                return -1;
            }
        } else {
            programs[vectorSize][1] = MakeProgram( device, source_private, sizeof(source_private) / sizeof( source_private[0]) );
            if( NULL == programs[ vectorSize ][1] )
            {
                gFailCount++;
                return -1;
            }
        }

        kernels[ vectorSize ][1] = clCreateKernel( programs[ vectorSize ][1], "test", &error );
        if( NULL == kernels[vectorSize][1] )
        {
            gFailCount++;
            vlog_error( "\t\tFAILED -- Failed to create private kernel. (%d)\n", error );
            return error;
        }

        if(g_arrVecSizes[vectorSize] == 3) {
            programs[vectorSize][2] = MakeProgram( device, source_local_v3, sizeof(source_local_v3) / sizeof( source_local_v3[0]) );
            if( NULL == programs[ vectorSize ][2] )
            {
                gFailCount++;
                return -1;
            }
        } else {
            programs[vectorSize][2] = MakeProgram( device, source_local, sizeof(source_local) / sizeof( source_local[0]) );
            if( NULL == programs[ vectorSize ][2] )
            {
                gFailCount++;
                return -1;
            }
        }

        kernels[ vectorSize ][2] = clCreateKernel( programs[ vectorSize ][2], "test", &error );
        if( NULL == kernels[vectorSize][2] )
        {
            gFailCount++;
            vlog_error( "\t\tFAILED -- Failed to create local kernel. (%d)\n", error );
            return error;
        }

        if( gTestDouble )
        {
            if(g_arrVecSizes[vectorSize] == 3) {
                doublePrograms[vectorSize][0] = MakeProgram( device, double_source_v3, sizeof(double_source_v3) / sizeof( double_source_v3[0]) );
                if( NULL == doublePrograms[ vectorSize ][0] )
                {
                    gFailCount++;
                    return -1;
                }
            } else {
                doublePrograms[vectorSize][0] = MakeProgram( device, double_source, sizeof(double_source) / sizeof( double_source[0]) );
                if( NULL == doublePrograms[ vectorSize ][0] )
                {
                    gFailCount++;
                    return -1;
                }
            }

            doubleKernels[ vectorSize ][0] = clCreateKernel( doublePrograms[ vectorSize ][0], "test", &error );
            if( NULL == kernels[vectorSize][0] )
            {
                gFailCount++;
                vlog_error( "\t\tFAILED -- Failed to create double kernel. (%d)\n", error );
                return error;
            }

            if(g_arrVecSizes[vectorSize] == 3) {
                doublePrograms[vectorSize][1] = MakeProgram( device, double_source_private_v3, sizeof(double_source_private_v3) / sizeof( double_source_private_v3[0]) );
                if( NULL == doublePrograms[ vectorSize ][1] )
                {
                    gFailCount++;
                    return -1;
                }
            } else {
                doublePrograms[vectorSize][1] = MakeProgram( device, double_source_private, sizeof(double_source_private) / sizeof( double_source_private[0]) );
                if( NULL == doublePrograms[ vectorSize ][1] )
                {
                    gFailCount++;
                    return -1;
                }
            }

            doubleKernels[ vectorSize ][1] = clCreateKernel( doublePrograms[ vectorSize ][1], "test", &error );
            if( NULL == kernels[vectorSize][1] )
            {
                gFailCount++;
                vlog_error( "\t\tFAILED -- Failed to create double private kernel. (%d)\n", error );
                return error;
            }

            if(g_arrVecSizes[vectorSize] == 3) {
                doublePrograms[vectorSize][2] = MakeProgram( device, double_source_local_v3, sizeof(double_source_local_v3) / sizeof( double_source_local_v3[0]) );
                if( NULL == doublePrograms[ vectorSize ][2] )
                {
                    gFailCount++;
                    return -1;
                }
            } else {
                doublePrograms[vectorSize][2] = MakeProgram( device, double_source_local, sizeof(double_source_local) / sizeof( double_source_local[0]) );
                if( NULL == doublePrograms[ vectorSize ][2] )
                {
                    gFailCount++;
                    return -1;
                }
            }

            doubleKernels[ vectorSize ][2] = clCreateKernel( doublePrograms[ vectorSize ][2], "test", &error );
            if( NULL == kernels[vectorSize][2] )
            {
                gFailCount++;
                vlog_error( "\t\tFAILED -- Failed to create double local kernel. (%d)\n", error );
                return error;
            }
        }
    }

    // Figure out how many elements are in a work block
    size_t elementSize = MAX( sizeof(cl_ushort), sizeof(float));
    size_t blockCount = BUFFER_SIZE / elementSize;
    uint64_t lastCase = 1ULL << (8*sizeof(float));
    size_t stride = blockCount;

    if (gWimpyMode)
        stride = (uint64_t)blockCount * (uint64_t)gWimpyReductionFactor;

    // we handle 64-bit types a bit differently.
    if( lastCase == 0 )
        lastCase = 0x100000000ULL;
    uint64_t i, j;
    error = 0;
    uint64_t printMask = (lastCase >> 4) - 1;
    cl_uint count = 0;
    int addressSpace;
    size_t loopCount;
    cl_uint threadCount = GetThreadCount();

    ComputeReferenceInfoF fref;
    fref.x = (float *)gIn_single;
    fref.r = (cl_half *)gOut_half_reference;
    fref.f = referenceFunc;
    fref.lim = blockCount;
    fref.count = (blockCount + threadCount - 1) / threadCount;

    CheckResultInfoF fchk;
    fchk.x = (const float *)gIn_single;
    fchk.r = (const cl_half *)gOut_half_reference;
    fchk.s = (const cl_half *)gOut_half;
    fchk.f = referenceFunc;
    fchk.lim = blockCount;
    fchk.count = (blockCount + threadCount - 1) / threadCount;

    ComputeReferenceInfoD dref;
    dref.x = (double *)gIn_double;
    dref.r = (cl_half *)gOut_half_reference_double;
    dref.f = doubleReferenceFunc;
    dref.lim = blockCount;
    dref.count = (blockCount + threadCount - 1) / threadCount;

    CheckResultInfoD dchk;
    dchk.x = (const double *)gIn_double;
    dchk.r = (const cl_half *)gOut_half_reference_double;
    dchk.s = (const cl_half *)gOut_half;
    dchk.f = doubleReferenceFunc;
    dchk.lim = blockCount;
    dchk.count = (blockCount + threadCount - 1) / threadCount;

    for( i = 0; i < (uint64_t)lastCase; i += stride )
    {
        count = (cl_uint) MIN( blockCount, lastCase - i );
        fref.i = i;
        dref.i = i;

        // Create the input and reference
        ThreadPool_Do(ReferenceF, threadCount, &fref);

        error = clEnqueueWriteBuffer(gQueue, gInBuffer_single, CL_FALSE, 0, count * sizeof(float ), gIn_single, 0, NULL, NULL);
        if (error) {
            vlog_error( "Failure in clWriteArray\n" );
            gFailCount++;
            goto exit;
        }

        if (gTestDouble) {
            ThreadPool_Do(ReferenceD, threadCount, &dref);

            error = clEnqueueWriteBuffer(gQueue, gInBuffer_double, CL_FALSE, 0, count * sizeof(double ), gIn_double, 0, NULL, NULL);
            if (error) {
                vlog_error( "Failure in clWriteArray\n" );
                gFailCount++;
                goto exit;
            }
        }

        for (vectorSize = minVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++) {
            // Loop over vector legths
            fchk.vsz = g_arrVecSizes[vectorSize];
            dchk.vsz = g_arrVecSizes[vectorSize];

            for ( addressSpace = 0; addressSpace < 3; addressSpace++) {
                // Loop over address spaces
                fchk.aspace = addressSpaceNames[addressSpace];
                dchk.aspace = addressSpaceNames[addressSpace];

                cl_uint pattern = 0xdeaddead;
                memset_pattern4(gOut_half, &pattern, BUFFER_SIZE/2);

                error = clEnqueueWriteBuffer(gQueue, gOutBuffer_half, CL_FALSE,
                                             0, count * sizeof(cl_half),
                                             gOut_half, 0, NULL, NULL);
                if (error) {
                    vlog_error( "Failure in clWriteArray\n" );
                    gFailCount++;
                    goto exit;
                }

                error = RunKernel(device, kernels[vectorSize][addressSpace], gInBuffer_single, gOutBuffer_half,
                                  numVecs(count, vectorSize, aligned),
                                  runsOverBy(count, vectorSize, aligned));
                if (error) {
                    gFailCount++;
                    goto exit;
                }

                error = clEnqueueReadBuffer(gQueue, gOutBuffer_half, CL_TRUE, 0,
                                            count * sizeof(cl_half), gOut_half,
                                            0, NULL, NULL);
                if (error) {
                    vlog_error( "Failure in clReadArray\n" );
                    gFailCount++;
                    goto exit;
                }

                error = ThreadPool_Do(CheckF, threadCount, &fchk);
                if (error) {
                            gFailCount++;
                            goto exit;
                        }

                if (gTestDouble) {
                    memset_pattern4(gOut_half, &pattern, BUFFER_SIZE/2);

                    error = clEnqueueWriteBuffer(
                        gQueue, gOutBuffer_half, CL_FALSE, 0,
                        count * sizeof(cl_half), gOut_half, 0, NULL, NULL);
                    if (error) {
                        vlog_error( "Failure in clWriteArray\n" );
                        gFailCount++;
                        goto exit;
                    }

                    error = RunKernel(device, doubleKernels[vectorSize][addressSpace], gInBuffer_double, gOutBuffer_half,
                                      numVecs(count, vectorSize, aligned),
                                      runsOverBy(count, vectorSize, aligned));
                    if (error) {
                        gFailCount++;
                        goto exit;
                    }

                    error = clEnqueueReadBuffer(
                        gQueue, gOutBuffer_half, CL_TRUE, 0,
                        count * sizeof(cl_half), gOut_half, 0, NULL, NULL);
                    if (error) {
                        vlog_error( "Failure in clReadArray\n" );
                        gFailCount++;
                        goto exit;
                    }

                    error = ThreadPool_Do(CheckD, threadCount, &dchk);
                    if (error) {
                                gFailCount++;
                                goto exit;
                            }
                        }
                    }
        }  // end for vector size

        if( ((i+blockCount) & ~printMask) == (i+blockCount) ) {
            vlog( "." );
            fflush( stdout );
        }
    }  // for end lastcase

    loopCount = count == blockCount ? 1 : 100;
    if( gReportTimes )
    {
        //Init the input stream
        cl_float *p = (cl_float *)gIn_single;
        for( j = 0; j < count; j++ )
            p[j] = (float)((double) (rand() - RAND_MAX/2) / (RAND_MAX/2));

        if( (error = clEnqueueWriteBuffer(gQueue, gInBuffer_single, CL_TRUE, 0, count * sizeof( float ), gIn_single, 0, NULL, NULL)) )
        {
            vlog_error( "Failure in clWriteArray\n" );
            gFailCount++;
            goto exit;
        }

        if( gTestDouble )
        {
            //Init the input stream
            cl_double *q = (cl_double *)gIn_double;
            for( j = 0; j < count; j++ )
                q[j] = ((double) (rand() - RAND_MAX/2) / (RAND_MAX/2));

            if( (error = clEnqueueWriteBuffer(gQueue, gInBuffer_double, CL_TRUE, 0, count * sizeof( double ), gIn_double, 0, NULL, NULL)) )
            {
                vlog_error( "Failure in clWriteArray\n" );
                gFailCount++;
                goto exit;
            }
        }

        //Run again for timing
        for( vectorSize = minVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
        {
            uint64_t bestTime = -1ULL;
            for( j = 0; j < loopCount; j++ )
            {
                uint64_t startTime = ReadTime();
                if( (error = RunKernel(device, kernels[vectorSize][0], gInBuffer_single, gOutBuffer_half, numVecs(count, vectorSize, aligned) ,
                                       runsOverBy(count, vectorSize, aligned)) ) )
                {
                    gFailCount++;
                    goto exit;
                }

                if( (error = clFinish(gQueue)) )
                {
                    vlog_error( "Failure in clFinish\n" );
                    gFailCount++;
                    goto exit;
                }
                uint64_t currentTime = ReadTime() - startTime;
                if( currentTime < bestTime )
                    bestTime = currentTime;
                time[ vectorSize ] += currentTime;
            }
            if( bestTime < min_time[ vectorSize ] )
                min_time[ vectorSize ] = bestTime ;

            if( gTestDouble )
            {
                bestTime = -1ULL;
                for( j = 0; j < loopCount; j++ )
                {
                    uint64_t startTime = ReadTime();
                    if( (error = RunKernel(device, doubleKernels[vectorSize][0], gInBuffer_double, gOutBuffer_half, numVecs(count, vectorSize, aligned) ,
                                           runsOverBy(count, vectorSize, aligned)) ) )
                    {
                        gFailCount++;
                        goto exit;
                    }

                    if( (error = clFinish(gQueue)) )
                    {
                        vlog_error( "Failure in clFinish\n" );
                        gFailCount++;
                        goto exit;
                    }
                    uint64_t currentTime = ReadTime() - startTime;
                    if( currentTime < bestTime )
                        bestTime = currentTime;
                    doubleTime[ vectorSize ] += currentTime;
                }
                if( bestTime < min_double_time[ vectorSize ] )
                    min_double_time[ vectorSize ] = bestTime;
            }
        }
    }

    if( gReportTimes )
    {
        for( vectorSize = minVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
            vlog_perf( SubtractTime( time[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) (count * loopCount), 0,
                      "average us/elem", "vStoreaHalf%s avg. (%s vector size: %d)", roundName, addressSpaceNames[0], (g_arrVecSizes[vectorSize]) );
        for( vectorSize = minVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
            vlog_perf( SubtractTime( min_time[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) count, 0,
                      "best us/elem", "vStoreaHalf%s best (%s vector size: %d)", roundName, addressSpaceNames[0], (g_arrVecSizes[vectorSize])  );
        if( gTestDouble )
        {
            for( vectorSize = minVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
                vlog_perf( SubtractTime( doubleTime[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) (count * loopCount), 0,
                          "average us/elem (double)", "vStoreaHalf%s avg. d (%s vector size: %d)", roundName, addressSpaceNames[0], (g_arrVecSizes[vectorSize])  );
            for( vectorSize = minVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
                vlog_perf( SubtractTime( min_double_time[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) count, 0,
                          "best us/elem (double)", "vStoreaHalf%s best d (%s vector size: %d)", roundName, addressSpaceNames[0], (g_arrVecSizes[vectorSize]) );
        }
    }

exit:
    //clean up
    for( vectorSize = minVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
    {
        for ( addressSpace = 0; addressSpace < 3; addressSpace++) {
            clReleaseKernel( kernels[ vectorSize ][addressSpace] );
            clReleaseProgram( programs[ vectorSize ][addressSpace] );
            if( gTestDouble )
            {
                clReleaseKernel( doubleKernels[ vectorSize ][addressSpace] );
                clReleaseProgram( doublePrograms[ vectorSize ][addressSpace] );
            }
        }
    }

    return error;
}