🏡 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 "testBase.h"
#include "harness/conversions.h"
#ifndef _WIN32
#include <unistd.h>
#endif

#define INT_TEST_VALUE 402258822
#define LONG_TEST_VALUE 515154531254381446LL


const char *atomic_global_pattern[] = {
    "__kernel void test_atomic_fn(volatile __global %s *destMemory, __global %s *oldValues)\n"
    "{\n"
    "    int  tid = get_global_id(0);\n"
    "\n"
    ,
    "\n"
    "}\n" };

const char *atomic_local_pattern[] = {
    "__kernel void test_atomic_fn(__global %s *finalDest, __global %s *oldValues, volatile __local %s *destMemory, int numDestItems )\n"
    "{\n"
    "    int  tid = get_global_id(0);\n"
    "     int  dstItemIdx;\n"
    "\n"
    "    // Everybody does the following line(s), but it all has the same result. We still need to ensure we sync before the atomic op, though\n"
    "     for( dstItemIdx = 0; dstItemIdx < numDestItems; dstItemIdx++ )\n"
    "        destMemory[ dstItemIdx ] = finalDest[ dstItemIdx ];\n"
    "    barrier( CLK_LOCAL_MEM_FENCE );\n"
    "\n"
    ,
    "    barrier( CLK_LOCAL_MEM_FENCE );\n"
    "    // Finally, write out the last value. Again, we're synced, so everyone will be writing the same value\n"
    "     for( dstItemIdx = 0; dstItemIdx < numDestItems; dstItemIdx++ )\n"
    "        finalDest[ dstItemIdx ] = destMemory[ dstItemIdx ];\n"
    "}\n" };


#define TEST_COUNT 128 * 1024


struct TestFns
{
    cl_int    mIntStartValue;
    cl_long    mLongStartValue;

    size_t    (*NumResultsFn)( size_t threadSize, ExplicitType dataType );

    // Integer versions
    cl_int    (*ExpectedValueIntFn)( size_t size, cl_int *startRefValues, size_t whichDestValue );
    void    (*GenerateRefsIntFn)( size_t size, cl_int *startRefValues, MTdata d );
    bool    (*VerifyRefsIntFn)( size_t size, cl_int *refValues, cl_int finalValue );

    // Long versions
    cl_long    (*ExpectedValueLongFn)( size_t size, cl_long *startRefValues, size_t whichDestValue );
    void    (*GenerateRefsLongFn)( size_t size, cl_long *startRefValues, MTdata d );
    bool    (*VerifyRefsLongFn)( size_t size, cl_long *refValues, cl_long finalValue );

    // Float versions
    cl_float    (*ExpectedValueFloatFn)( size_t size, cl_float *startRefValues, size_t whichDestValue );
    void        (*GenerateRefsFloatFn)( size_t size, cl_float *startRefValues, MTdata d );
    bool        (*VerifyRefsFloatFn)( size_t size, cl_float *refValues, cl_float finalValue );
};

bool check_atomic_support( cl_device_id device, bool extended, bool isLocal, ExplicitType dataType )
{
    const char *extensionNames[8] = {
        "cl_khr_global_int32_base_atomics", "cl_khr_global_int32_extended_atomics",
        "cl_khr_local_int32_base_atomics",  "cl_khr_local_int32_extended_atomics",
        "cl_khr_int64_base_atomics",        "cl_khr_int64_extended_atomics",
        "cl_khr_int64_base_atomics",        "cl_khr_int64_extended_atomics"       // this line intended to be the same as the last one
    };

    size_t index = 0;
    if( extended )
        index += 1;
    if( isLocal )
        index += 2;

    Version version = get_device_cl_version(device);

    switch (dataType)
    {
        case kInt:
        case kUInt:
            if( version >= Version(1,1) )
                return 1;
            break;
        case kLong:
        case kULong:
            index += 4;
            break;
        case kFloat:  // this has to stay separate since the float atomics arent in the 1.0 extensions
            return version >= Version(1,1);
        default:
            log_error( "ERROR:  Unsupported data type (%d) in check_atomic_support\n", dataType );
            return 0;
    }

    return is_extension_available( device, extensionNames[index] );
}

int test_atomic_function(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements, const char *programCore,
                         TestFns testFns,
                         bool extended, bool isLocal, ExplicitType dataType, bool matchGroupSize )
{
    clProgramWrapper program;
    clKernelWrapper kernel;
    int error;
    size_t threads[1];
    clMemWrapper streams[2];
    void *refValues, *startRefValues;
    size_t threadSize, groupSize;
    const char *programLines[4];
    char pragma[ 512 ];
    char programHeader[ 512 ];
    MTdata d;
    size_t typeSize = get_explicit_type_size( dataType );


    // Verify we can run first
    bool isUnsigned = ( dataType == kULong ) || ( dataType == kUInt );
    if( !check_atomic_support( deviceID, extended, isLocal, dataType ) )
    {
        // Only print for the signed (unsigned comes right after, and if signed isn't supported, unsigned isn't either)
        if( dataType == kFloat )
            log_info( "\t%s float not supported\n", isLocal ? "Local" : "Global" );
        else if( !isUnsigned )
            log_info( "\t%s %sint%d not supported\n", isLocal ? "Local" : "Global", isUnsigned ? "u" : "", (int)typeSize * 8 );
        // Since we don't support the operation, they implicitly pass
        return 0;
    }
    else
    {
        if( dataType == kFloat )
            log_info( "\t%s float%s...", isLocal ? "local" : "global", isLocal ? " " : "" );
        else
            log_info( "\t%s %sint%d%s%s...", isLocal ? "local" : "global", isUnsigned ? "u" : "",
                     (int)typeSize * 8, isUnsigned ? "" : " ", isLocal ? " " : "" );
    }

    //// Set up the kernel code

    // Create the pragma line for this kernel
    bool isLong = ( dataType == kLong || dataType == kULong );
    sprintf( pragma, "#pragma OPENCL EXTENSION cl_khr%s_int%s_%s_atomics : enable\n",
            isLong ? "" : (isLocal ? "_local" : "_global"), isLong ? "64" : "32",
            extended ? "extended" : "base" );

    // Now create the program header
    const char *typeName = get_explicit_type_name( dataType );
    if( isLocal )
        sprintf( programHeader, atomic_local_pattern[ 0 ], typeName, typeName, typeName );
    else
        sprintf( programHeader, atomic_global_pattern[ 0 ], typeName, typeName );

    // Set up our entire program now
    programLines[ 0 ] = pragma;
    programLines[ 1 ] = programHeader;
    programLines[ 2 ] = programCore;
    programLines[ 3 ] = ( isLocal ) ? atomic_local_pattern[ 1 ] : atomic_global_pattern[ 1 ];

    if( create_single_kernel_helper( context, &program, &kernel, 4, programLines, "test_atomic_fn" ) )
    {
        return -1;
    }

    //// Set up to actually run
    threadSize = num_elements;

    error = get_max_common_work_group_size( context, kernel, threadSize, &groupSize );
    test_error( error, "Unable to get thread group max size" );

    if( matchGroupSize )
        // HACK because xchg and cmpxchg apparently are limited by hardware
        threadSize = groupSize;

    if( isLocal )
    {
        size_t maxSizes[3] = {0, 0, 0};
        error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WORK_ITEM_SIZES, 3*sizeof(size_t), maxSizes, 0);
        test_error( error, "Unable to obtain max work item sizes for the device" );

        size_t workSize;
        error = clGetKernelWorkGroupInfo( kernel, deviceID, CL_KERNEL_WORK_GROUP_SIZE, sizeof( workSize ), &workSize, NULL );
        test_error( error, "Unable to obtain max work group size for device and kernel combo" );

        // "workSize" is limited to that of the first dimension as only a 1DRange is executed.
        if( maxSizes[0] < workSize )
        {
            workSize = maxSizes[0];
        }

        threadSize = groupSize = workSize;
    }


    log_info( "\t(thread count %d, group size %d)\n", (int)threadSize, (int)groupSize );

    refValues = (cl_int *)malloc( typeSize * threadSize );

    if( testFns.GenerateRefsIntFn != NULL )
    {
        // We have a ref generator provided
        d = init_genrand( gRandomSeed );
        startRefValues = malloc( typeSize * threadSize );
        if( typeSize == 4 )
            testFns.GenerateRefsIntFn( threadSize, (cl_int *)startRefValues, d );
        else
            testFns.GenerateRefsLongFn( threadSize, (cl_long *)startRefValues, d );
        free_mtdata(d);
        d = NULL;
    }
    else
        startRefValues = NULL;

    // If we're given a num_results function, we need to determine how many result objects we need. If
    // we don't have it, we assume it's just 1
    size_t numDestItems = ( testFns.NumResultsFn != NULL ) ? testFns.NumResultsFn( threadSize, dataType ) : 1;

    char * destItems = new char[ typeSize * numDestItems ];
    if( destItems == NULL )
    {
        log_error( "ERROR: Unable to allocate memory!\n" );
        return -1;
    }
    void * startValue = ( typeSize == 4 ) ? (void *)&testFns.mIntStartValue : (void *)&testFns.mLongStartValue;
    for( size_t i = 0; i < numDestItems; i++ )
        memcpy( destItems + i * typeSize, startValue, typeSize );

    streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
                                typeSize * numDestItems, destItems, NULL);
    if (!streams[0])
    {
        log_error("ERROR: Creating output array failed!\n");
        return -1;
    }
    streams[1] = clCreateBuffer(
        context,
        ((startRefValues != NULL ? CL_MEM_COPY_HOST_PTR : CL_MEM_READ_WRITE)),
        typeSize * threadSize, startRefValues, NULL);
    if (!streams[1])
    {
        log_error("ERROR: Creating reference array failed!\n");
        return -1;
    }

    /* Set the arguments */
    error = clSetKernelArg( kernel, 0, sizeof( streams[0] ), &streams[0] );
    test_error( error, "Unable to set indexed kernel arguments" );
    error = clSetKernelArg( kernel, 1, sizeof( streams[1] ), &streams[1] );
    test_error( error, "Unable to set indexed kernel arguments" );

    if( isLocal )
    {
        error = clSetKernelArg( kernel, 2, typeSize * numDestItems, NULL );
        test_error( error, "Unable to set indexed local kernel argument" );

        cl_int numDestItemsInt = (cl_int)numDestItems;
        error = clSetKernelArg( kernel, 3, sizeof( cl_int ), &numDestItemsInt );
        test_error( error, "Unable to set indexed kernel argument" );
    }

    /* Run the kernel */
    threads[0] = threadSize;
    error = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, threads, &groupSize, 0, NULL, NULL );
    test_error( error, "Unable to execute test kernel" );

    error = clEnqueueReadBuffer( queue, streams[0], true, 0, typeSize * numDestItems, destItems, 0, NULL, NULL );
    test_error( error, "Unable to read result value!" );

    error = clEnqueueReadBuffer( queue, streams[1], true, 0, typeSize * threadSize, refValues, 0, NULL, NULL );
    test_error( error, "Unable to read reference values!" );

    // If we have an expectedFn, then we need to generate a final value to compare against. If we don't
    // have one, it's because we're comparing ref values only
    if( testFns.ExpectedValueIntFn != NULL )
    {
        for( size_t i = 0; i < numDestItems; i++ )
        {
            char expected[ 8 ];
            cl_int intVal;
            cl_long longVal;
            if( typeSize == 4 )
            {
                // Int version
                intVal = testFns.ExpectedValueIntFn( threadSize, (cl_int *)startRefValues, i );
                memcpy( expected, &intVal, sizeof( intVal ) );
            }
            else
            {
                // Long version
                longVal = testFns.ExpectedValueLongFn( threadSize, (cl_long *)startRefValues, i );
                memcpy( expected, &longVal, sizeof( longVal ) );
            }

            if( memcmp( expected, destItems + i * typeSize, typeSize ) != 0 )
            {
                if( typeSize == 4 )
                {
                    cl_int *outValue = (cl_int *)( destItems + i * typeSize );
                    log_error( "ERROR: Result %ld from kernel does not validate! (should be %d, was %d)\n", i, intVal, *outValue );
                    cl_int *startRefs = (cl_int *)startRefValues;
                    cl_int *refs = (cl_int *)refValues;
                    for( i = 0; i < threadSize; i++ )
                    {
                        if( startRefs != NULL )
                            log_info( " --- %ld - %d --- %d\n", i, startRefs[i], refs[i] );
                        else
                            log_info( " --- %ld --- %d\n", i, refs[i] );
                    }
                }
                else
                {
                    cl_long *outValue = (cl_long *)( destItems + i * typeSize );
                    log_error( "ERROR: Result %ld from kernel does not validate! (should be %lld, was %lld)\n", i, longVal, *outValue );
                    cl_long *startRefs = (cl_long *)startRefValues;
                    cl_long *refs = (cl_long *)refValues;
                    for( i = 0; i < threadSize; i++ )
                    {
                        if( startRefs != NULL )
                            log_info( " --- %ld - %lld --- %lld\n", i, startRefs[i], refs[i] );
                        else
                            log_info( " --- %ld --- %lld\n", i, refs[i] );
                    }
                }
                return -1;
            }
        }
    }

    if( testFns.VerifyRefsIntFn != NULL )
    {
        /* Use the verify function to also check the results */
        if( dataType == kFloat )
        {
            cl_float *outValue = (cl_float *)destItems;
            if( !testFns.VerifyRefsFloatFn( threadSize, (cl_float *)refValues, *outValue ) != 0 )
            {
                log_error( "ERROR: Reference values did not validate!\n" );
                return -1;
            }
        }
        else if( typeSize == 4 )
        {
            cl_int *outValue = (cl_int *)destItems;
            if( !testFns.VerifyRefsIntFn( threadSize, (cl_int *)refValues, *outValue ) != 0 )
            {
                log_error( "ERROR: Reference values did not validate!\n" );
                return -1;
            }
        }
        else
        {
            cl_long *outValue = (cl_long *)destItems;
            if( !testFns.VerifyRefsLongFn( threadSize, (cl_long *)refValues, *outValue ) != 0 )
            {
                log_error( "ERROR: Reference values did not validate!\n" );
                return -1;
            }
        }
    }
    else if( testFns.ExpectedValueIntFn == NULL )
    {
        log_error( "ERROR: Test doesn't check total or refs; no values are verified!\n" );
        return -1;
    }


    /* Re-write the starting value */
    for( size_t i = 0; i < numDestItems; i++ )
        memcpy( destItems + i * typeSize, startValue, typeSize );
    error = clEnqueueWriteBuffer( queue, streams[0], true, 0, typeSize * numDestItems, destItems, 0, NULL, NULL );
    test_error( error, "Unable to write starting values!" );

    /* Run the kernel once for a single thread, so we can verify that the returned value is the original one */
    threads[0] = 1;
    error = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, threads, threads, 0, NULL, NULL );
    test_error( error, "Unable to execute test kernel" );

    error = clEnqueueReadBuffer( queue, streams[1], true, 0, typeSize, refValues, 0, NULL, NULL );
    test_error( error, "Unable to read reference values!" );

    if( memcmp( refValues, destItems, typeSize ) != 0 )
    {
        if( typeSize == 4 )
        {
            cl_int *s = (cl_int *)destItems;
            cl_int *r = (cl_int *)refValues;
            log_error( "ERROR: atomic function operated correctly but did NOT return correct 'old' value "
                      " (should have been %d, returned %d)!\n", *s, *r );
        }
        else
        {
            cl_long *s = (cl_long *)destItems;
            cl_long *r = (cl_long *)refValues;
            log_error( "ERROR: atomic function operated correctly but did NOT return correct 'old' value "
                      " (should have been %lld, returned %lld)!\n", *s, *r );
        }
        return -1;
    }

    delete [] destItems;
    free( refValues );
    if( startRefValues != NULL )
        free( startRefValues );

    return 0;
}

int test_atomic_function_set(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements, const char *programCore,
                             TestFns testFns,
                             bool extended, bool matchGroupSize, bool usingAtomicPrefix )
{
    log_info("    Testing %s functions...\n", usingAtomicPrefix ? "atomic_" : "atom_");

    int errors = 0;
    errors |= test_atomic_function( deviceID, context, queue, num_elements, programCore, testFns, extended, false, kInt, matchGroupSize );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, programCore, testFns, extended, false, kUInt, matchGroupSize );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, programCore, testFns, extended, true, kInt, matchGroupSize );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, programCore, testFns, extended, true, kUInt, matchGroupSize );

    // Only the 32 bit atomic functions use the "atomic" prefix in 1.1, the 64 bit functions still use the "atom" prefix.
    // The argument usingAtomicPrefix is set to true if programCore was generated with the "atomic" prefix.
    if (!usingAtomicPrefix) {
      errors |= test_atomic_function( deviceID, context, queue, num_elements, programCore, testFns, extended, false, kLong, matchGroupSize );
      errors |= test_atomic_function( deviceID, context, queue, num_elements, programCore, testFns, extended, false, kULong, matchGroupSize );
      errors |= test_atomic_function( deviceID, context, queue, num_elements, programCore, testFns, extended, true, kLong, matchGroupSize );
      errors |= test_atomic_function( deviceID, context, queue, num_elements, programCore, testFns, extended, true, kULong, matchGroupSize );
    }

    return errors;
}

#pragma mark ---- add

const char atom_add_core[] =
"    oldValues[tid] = atom_add( &destMemory[0], tid + 3 );\n"
"    atom_add( &destMemory[0], tid + 3 );\n"
"   atom_add( &destMemory[0], tid + 3 );\n"
"   atom_add( &destMemory[0], tid + 3 );\n";

const char atomic_add_core[] =
"    oldValues[tid] = atomic_add( &destMemory[0], tid + 3 );\n"
"    atomic_add( &destMemory[0], tid + 3 );\n"
"   atomic_add( &destMemory[0], tid + 3 );\n"
"   atomic_add( &destMemory[0], tid + 3 );\n";

cl_int test_atomic_add_result_int( size_t size, cl_int *startRefValues, size_t whichDestValue )
{
    cl_int total = 0;
    for( size_t i = 0; i < size; i++ )
        total += ( (cl_int)i + 3 ) * 4;
    return total;
}

cl_long test_atomic_add_result_long( size_t size, cl_long *startRefValues, size_t whichDestValue )
{
    cl_long total = 0;
    for( size_t i = 0; i < size; i++ )
        total += ( ( i + 3 ) * 4 );
    return total;
}

int test_atomic_add(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { 0, 0LL, NULL, test_atomic_add_result_int, NULL, NULL, test_atomic_add_result_long, NULL, NULL };

    if( test_atomic_function_set( deviceID, context, queue, num_elements, atom_add_core, set, false, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ false ) != 0 )
        return -1;
    if( test_atomic_function_set( deviceID, context, queue, num_elements, atomic_add_core, set, false, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ true ) != 0 )
      return -1;
    return 0;
}

#pragma mark ---- sub

const char atom_sub_core[] = "    oldValues[tid] = atom_sub( &destMemory[0], tid + 3 );\n";

const char atomic_sub_core[] = "    oldValues[tid] = atomic_sub( &destMemory[0], tid + 3 );\n";

cl_int test_atomic_sub_result_int( size_t size, cl_int *startRefValues, size_t whichDestValue )
{
    cl_int total = INT_TEST_VALUE;
    for( size_t i = 0; i < size; i++ )
        total -= (cl_int)i + 3;
    return total;
}

cl_long test_atomic_sub_result_long( size_t size, cl_long *startRefValues, size_t whichDestValue )
{
    cl_long total = LONG_TEST_VALUE;
    for( size_t i = 0; i < size; i++ )
        total -= i + 3;
    return total;
}

int test_atomic_sub(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { INT_TEST_VALUE, LONG_TEST_VALUE, NULL, test_atomic_sub_result_int, NULL, NULL, test_atomic_sub_result_long, NULL, NULL };

    if( test_atomic_function_set( deviceID, context, queue, num_elements, atom_sub_core, set, false, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ false  ) != 0 )
        return -1;
    if( test_atomic_function_set( deviceID, context, queue, num_elements, atomic_sub_core, set, false, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ true  ) != 0 )
        return -1;
    return 0;
}

#pragma mark ---- xchg

const char atom_xchg_core[] = "    oldValues[tid] = atom_xchg( &destMemory[0], tid );\n";

const char atomic_xchg_core[] = "    oldValues[tid] = atomic_xchg( &destMemory[0], tid );\n";
const char atomic_xchg_float_core[] = "    oldValues[tid] = atomic_xchg( &destMemory[0], tid );\n";

bool test_atomic_xchg_verify_int( size_t size, cl_int *refValues, cl_int finalValue )
{
    /* For xchg, each value from 0 to size - 1 should have an entry in the ref array, and ONLY one entry */
    char *valids;
    size_t i;
    char originalValidCount = 0;

    valids = (char *)malloc( sizeof( char ) * size );
    memset( valids, 0, sizeof( char ) * size );

    for( i = 0; i < size; i++ )
    {
        if( refValues[ i ] == INT_TEST_VALUE )
        {
            // Special initial value
            originalValidCount++;
            continue;
        }
        if( refValues[ i ] < 0 || (size_t)refValues[ i ] >= size )
        {
            log_error( "ERROR: Reference value %ld outside of valid range! (%d)\n", i, refValues[ i ] );
            return false;
        }
        valids[ refValues[ i ] ] ++;
    }

    /* Note: ONE entry will have zero count. It'll be the last one that executed, because that value should be
     the final value outputted */
    if( valids[ finalValue ] > 0 )
    {
        log_error( "ERROR: Final value %d was also in ref list!\n", finalValue );
        return false;
    }
    else
        valids[ finalValue ] = 1;    // So the following loop will be okay

    /* Now check that every entry has one and only one count */
    if( originalValidCount != 1 )
    {
        log_error( "ERROR: Starting reference value %d did not occur once-and-only-once (occurred %d)\n", 65191, originalValidCount );
        return false;
    }
    for( i = 0; i < size; i++ )
    {
        if( valids[ i ] != 1 )
        {
            log_error( "ERROR: Reference value %ld did not occur once-and-only-once (occurred %d)\n", i, valids[ i ] );
            for( size_t j = 0; j < size; j++ )
                log_info( "%d: %d\n", (int)j, (int)valids[ j ] );
            return false;
        }
    }

    free( valids );
    return true;
}

bool test_atomic_xchg_verify_long( size_t size, cl_long *refValues, cl_long finalValue )
{
    /* For xchg, each value from 0 to size - 1 should have an entry in the ref array, and ONLY one entry */
    char *valids;
    size_t i;
    char originalValidCount = 0;

    valids = (char *)malloc( sizeof( char ) * size );
    memset( valids, 0, sizeof( char ) * size );

    for( i = 0; i < size; i++ )
    {
        if( refValues[ i ] == LONG_TEST_VALUE )
        {
            // Special initial value
            originalValidCount++;
            continue;
        }
        if( refValues[ i ] < 0 || (size_t)refValues[ i ] >= size )
        {
            log_error( "ERROR: Reference value %ld outside of valid range! (%lld)\n", i, refValues[ i ] );
            return false;
        }
        valids[ refValues[ i ] ] ++;
    }

    /* Note: ONE entry will have zero count. It'll be the last one that executed, because that value should be
     the final value outputted */
    if( valids[ finalValue ] > 0 )
    {
        log_error( "ERROR: Final value %lld was also in ref list!\n", finalValue );
        return false;
    }
    else
        valids[ finalValue ] = 1;    // So the following loop will be okay

    /* Now check that every entry has one and only one count */
    if( originalValidCount != 1 )
    {
        log_error( "ERROR: Starting reference value %d did not occur once-and-only-once (occurred %d)\n", 65191, originalValidCount );
        return false;
    }
    for( i = 0; i < size; i++ )
    {
        if( valids[ i ] != 1 )
        {
            log_error( "ERROR: Reference value %ld did not occur once-and-only-once (occurred %d)\n", i, valids[ i ] );
            for( size_t j = 0; j < size; j++ )
                log_info( "%d: %d\n", (int)j, (int)valids[ j ] );
            return false;
        }
    }

    free( valids );
    return true;
}

bool test_atomic_xchg_verify_float( size_t size, cl_float *refValues, cl_float finalValue )
{
    /* For xchg, each value from 0 to size - 1 should have an entry in the ref array, and ONLY one entry */
    char *valids;
    size_t i;
    char originalValidCount = 0;

    valids = (char *)malloc( sizeof( char ) * size );
    memset( valids, 0, sizeof( char ) * size );

    for( i = 0; i < size; i++ )
    {
        cl_int *intRefValue = (cl_int *)( &refValues[ i ] );
        if( *intRefValue == INT_TEST_VALUE )
        {
            // Special initial value
            originalValidCount++;
            continue;
        }
        if( refValues[ i ] < 0 || (size_t)refValues[ i ] >= size )
        {
            log_error( "ERROR: Reference value %ld outside of valid range! (%a)\n", i, refValues[ i ] );
            return false;
        }
        valids[ (int)refValues[ i ] ] ++;
    }

    /* Note: ONE entry will have zero count. It'll be the last one that executed, because that value should be
     the final value outputted */
    if( valids[ (int)finalValue ] > 0 )
    {
        log_error( "ERROR: Final value %a was also in ref list!\n", finalValue );
        return false;
    }
    else
        valids[ (int)finalValue ] = 1;    // So the following loop will be okay

    /* Now check that every entry has one and only one count */
    if( originalValidCount != 1 )
    {
        log_error( "ERROR: Starting reference value %d did not occur once-and-only-once (occurred %d)\n", 65191, originalValidCount );
        return false;
    }
    for( i = 0; i < size; i++ )
    {
        if( valids[ i ] != 1 )
        {
            log_error( "ERROR: Reference value %ld did not occur once-and-only-once (occurred %d)\n", i, valids[ i ] );
            for( size_t j = 0; j < size; j++ )
                log_info( "%d: %d\n", (int)j, (int)valids[ j ] );
            return false;
        }
    }

    free( valids );
    return true;
}

int test_atomic_xchg(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { INT_TEST_VALUE, LONG_TEST_VALUE, NULL, NULL, NULL, test_atomic_xchg_verify_int, NULL, NULL, test_atomic_xchg_verify_long, NULL, NULL, test_atomic_xchg_verify_float };

    int errors = test_atomic_function_set( deviceID, context, queue, num_elements, atom_xchg_core, set, false, true, /*usingAtomicPrefix*/ false  );
    errors |= test_atomic_function_set( deviceID, context, queue, num_elements, atomic_xchg_core, set, false, true, /*usingAtomicPrefix*/ true  );

    errors |= test_atomic_function( deviceID, context, queue, num_elements, atomic_xchg_float_core, set, false, false, kFloat, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atomic_xchg_float_core, set, false, true, kFloat, true );

    return errors;
}


#pragma mark ---- min

const char atom_min_core[] = "    oldValues[tid] = atom_min( &destMemory[0], oldValues[tid] );\n";

const char atomic_min_core[] = "    oldValues[tid] = atomic_min( &destMemory[0], oldValues[tid] );\n";

cl_int test_atomic_min_result_int( size_t size, cl_int *startRefValues, size_t whichDestValue )
{
    cl_int total = 0x7fffffffL;
    for( size_t i = 0; i < size; i++ )
    {
        if( startRefValues[ i ] < total )
            total = startRefValues[ i ];
    }
    return total;
}

void test_atomic_min_gen_int( size_t size, cl_int *startRefValues, MTdata d )
{
    for( size_t i = 0; i < size; i++ )
        startRefValues[i] = (cl_int)( genrand_int32(d) % 0x3fffffff ) + 0x3fffffff;
}

cl_long test_atomic_min_result_long( size_t size, cl_long *startRefValues, size_t whichDestValue )
{
    cl_long total = 0x7fffffffffffffffLL;
    for( size_t i = 0; i < size; i++ )
    {
        if( startRefValues[ i ] < total )
            total = startRefValues[ i ];
    }
    return total;
}

void test_atomic_min_gen_long( size_t size, cl_long *startRefValues, MTdata d )
{
    for( size_t i = 0; i < size; i++ )
        startRefValues[i] = (cl_long)( genrand_int32(d) | ( ( (cl_long)genrand_int32(d) & 0x7fffffffL ) << 16 ) );
}

int test_atomic_min(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { 0x7fffffffL, 0x7fffffffffffffffLL, NULL, test_atomic_min_result_int, test_atomic_min_gen_int, NULL, test_atomic_min_result_long, test_atomic_min_gen_long, NULL };

    if( test_atomic_function_set( deviceID, context, queue, num_elements, atom_min_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ false  ) != 0 )
        return -1;
    if( test_atomic_function_set( deviceID, context, queue, num_elements, atomic_min_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ true  ) != 0 )
        return -1;
    return 0;
}


#pragma mark ---- max

const char atom_max_core[] = "    oldValues[tid] = atom_max( &destMemory[0], oldValues[tid] );\n";

const char atomic_max_core[] = "    oldValues[tid] = atomic_max( &destMemory[0], oldValues[tid] );\n";

cl_int test_atomic_max_result_int( size_t size, cl_int *startRefValues, size_t whichDestValue )
{
    cl_int total = 0;
    for( size_t i = 0; i < size; i++ )
    {
        if( startRefValues[ i ] > total )
            total = startRefValues[ i ];
    }
    return total;
}

void test_atomic_max_gen_int( size_t size, cl_int *startRefValues, MTdata d )
{
    for( size_t i = 0; i < size; i++ )
        startRefValues[i] = (cl_int)( genrand_int32(d) % 0x3fffffff ) + 0x3fffffff;
}

cl_long test_atomic_max_result_long( size_t size, cl_long *startRefValues, size_t whichDestValue )
{
    cl_long total = 0;
    for( size_t i = 0; i < size; i++ )
    {
        if( startRefValues[ i ] > total )
            total = startRefValues[ i ];
    }
    return total;
}

void test_atomic_max_gen_long( size_t size, cl_long *startRefValues, MTdata d )
{
    for( size_t i = 0; i < size; i++ )
        startRefValues[i] = (cl_long)( genrand_int32(d) | ( ( (cl_long)genrand_int32(d) & 0x7fffffffL ) << 16 ) );
}

int test_atomic_max(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { 0, 0, NULL, test_atomic_max_result_int, test_atomic_max_gen_int, NULL, test_atomic_max_result_long, test_atomic_max_gen_long, NULL };

    if( test_atomic_function_set( deviceID, context, queue, num_elements, atom_max_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ false  ) != 0 )
        return -1;
    if( test_atomic_function_set( deviceID, context, queue, num_elements, atomic_max_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ true  ) != 0 )
      return -1;
    return 0;
}


#pragma mark ---- inc

const char atom_inc_core[] = "    oldValues[tid] = atom_inc( &destMemory[0] );\n";

const char atomic_inc_core[] = "    oldValues[tid] = atomic_inc( &destMemory[0] );\n";

cl_int test_atomic_inc_result_int( size_t size, cl_int *startRefValues, size_t whichDestValue )
{
    return INT_TEST_VALUE + (cl_int)size;
}

cl_long test_atomic_inc_result_long( size_t size, cl_long *startRefValues, size_t whichDestValue )
{
    return LONG_TEST_VALUE + size;
}

int test_atomic_inc(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { INT_TEST_VALUE, LONG_TEST_VALUE, NULL, test_atomic_inc_result_int, NULL, NULL, test_atomic_inc_result_long, NULL, NULL };

    if( test_atomic_function_set( deviceID, context, queue, num_elements, atom_inc_core, set, false, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ false  ) != 0 )
        return -1;
    if( test_atomic_function_set( deviceID, context, queue, num_elements, atomic_inc_core, set, false, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ true  ) != 0 )
        return -1;
    return 0;
}


#pragma mark ---- dec

const char atom_dec_core[] = "    oldValues[tid] = atom_dec( &destMemory[0] );\n";

const char atomic_dec_core[] = "    oldValues[tid] = atomic_dec( &destMemory[0] );\n";

cl_int test_atomic_dec_result_int( size_t size, cl_int *startRefValues, size_t whichDestValue )
{
    return INT_TEST_VALUE - (cl_int)size;
}

cl_long test_atomic_dec_result_long( size_t size, cl_long *startRefValues, size_t whichDestValue )
{
    return LONG_TEST_VALUE - size;
}

int test_atomic_dec(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { INT_TEST_VALUE, LONG_TEST_VALUE, NULL, test_atomic_dec_result_int, NULL, NULL, test_atomic_dec_result_long, NULL, NULL };

    if( test_atomic_function_set( deviceID, context, queue, num_elements, atom_dec_core, set, false, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ false  ) != 0 )
        return -1;
    if( test_atomic_function_set( deviceID, context, queue, num_elements, atomic_dec_core, set, false, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ true  ) != 0 )
        return -1;
    return 0;
}


#pragma mark ---- cmpxchg

/* We test cmpxchg by implementing (the long way) atom_add */
const char atom_cmpxchg_core[] =
"    int oldValue, origValue, newValue;\n"
"    do { \n"
"        origValue = destMemory[0];\n"
"        newValue = origValue + tid + 2;\n"
"        oldValue = atom_cmpxchg( &destMemory[0], origValue, newValue );\n"
"    } while( oldValue != origValue );\n"
"    oldValues[tid] = oldValue;\n"
;

const char atom_cmpxchg64_core[] =
"    long oldValue, origValue, newValue;\n"
"    do { \n"
"        origValue = destMemory[0];\n"
"        newValue = origValue + tid + 2;\n"
"        oldValue = atom_cmpxchg( &destMemory[0], origValue, newValue );\n"
"    } while( oldValue != origValue );\n"
"    oldValues[tid] = oldValue;\n"
;

const char atomic_cmpxchg_core[] =
"    int oldValue, origValue, newValue;\n"
"    do { \n"
"        origValue = destMemory[0];\n"
"        newValue = origValue + tid + 2;\n"
"        oldValue = atomic_cmpxchg( &destMemory[0], origValue, newValue );\n"
"    } while( oldValue != origValue );\n"
"    oldValues[tid] = oldValue;\n"
;

cl_int test_atomic_cmpxchg_result_int( size_t size, cl_int *startRefValues, size_t whichDestValue )
{
    cl_int total = INT_TEST_VALUE;
    for( size_t i = 0; i < size; i++ )
        total += (cl_int)i + 2;
    return total;
}

cl_long test_atomic_cmpxchg_result_long( size_t size, cl_long *startRefValues, size_t whichDestValue )
{
    cl_long total = LONG_TEST_VALUE;
    for( size_t i = 0; i < size; i++ )
        total += i + 2;
    return total;
}

int test_atomic_cmpxchg(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { INT_TEST_VALUE, LONG_TEST_VALUE, NULL, test_atomic_cmpxchg_result_int, NULL, NULL, test_atomic_cmpxchg_result_long, NULL, NULL };

    int errors = 0;

    log_info("    Testing atom_ functions...\n");
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atom_cmpxchg_core, set, false, false, kInt, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atom_cmpxchg_core, set, false, false, kUInt, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atom_cmpxchg_core, set, false, true, kInt, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atom_cmpxchg_core, set, false, true, kUInt, true );

    errors |= test_atomic_function( deviceID, context, queue, num_elements, atom_cmpxchg64_core, set, false, false, kLong, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atom_cmpxchg64_core, set, false, false, kULong, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atom_cmpxchg64_core, set, false, true, kLong, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atom_cmpxchg64_core, set, false, true, kULong, true );

    log_info("    Testing atomic_ functions...\n");
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atomic_cmpxchg_core, set, false, false, kInt, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atomic_cmpxchg_core, set, false, false, kUInt, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atomic_cmpxchg_core, set, false, true, kInt, true );
    errors |= test_atomic_function( deviceID, context, queue, num_elements, atomic_cmpxchg_core, set, false, true, kUInt, true );

    if( errors )
        return -1;

    return 0;
}

#pragma mark -------- Bitwise functions

size_t test_bitwise_num_results( size_t threadCount, ExplicitType dataType )
{
    size_t numBits = get_explicit_type_size( dataType ) * 8;

    return ( threadCount + numBits - 1 ) / numBits;
}

#pragma mark ---- and

const char atom_and_core[] =
"    size_t numBits = sizeof( destMemory[0] ) * 8;\n"
"    int  whichResult = tid / numBits;\n"
"    int  bitIndex = tid - ( whichResult * numBits );\n"
"\n"
"    oldValues[tid] = atom_and( &destMemory[whichResult], ~( 1L << bitIndex ) );\n"
;

const char atomic_and_core[] =
"    size_t numBits = sizeof( destMemory[0] ) * 8;\n"
"    int  whichResult = tid / numBits;\n"
"    int  bitIndex = tid - ( whichResult * numBits );\n"
"\n"
"    oldValues[tid] = atomic_and( &destMemory[whichResult], ~( 1L << bitIndex ) );\n"
;


cl_int test_atomic_and_result_int( size_t size, cl_int *startRefValues, size_t whichResult )
{
    size_t numThreads = ( (size_t)size + 31 ) / 32;
    if( whichResult < numThreads - 1 )
        return 0;

    // Last item doesn't get and'ed on every bit, so we have to mask away
    size_t numBits = (size_t)size - whichResult * 32;
    cl_int bits = (cl_int)0xffffffffL;
    for( size_t i = 0; i < numBits; i++ )
        bits &= ~( 1 << i );

    return bits;
}

cl_long test_atomic_and_result_long( size_t size, cl_long *startRefValues, size_t whichResult )
{
    size_t numThreads = ( (size_t)size + 63 ) / 64;
    if( whichResult < numThreads - 1 )
        return 0;

    // Last item doesn't get and'ed on every bit, so we have to mask away
    size_t numBits = (size_t)size - whichResult * 64;
    cl_long bits = (cl_long)0xffffffffffffffffLL;
    for (size_t i = 0; i < numBits; i++) bits &= ~(1LL << i);

    return bits;
}

int test_atomic_and(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { 0xffffffff, 0xffffffffffffffffLL, test_bitwise_num_results,
        test_atomic_and_result_int, NULL, NULL, test_atomic_and_result_long, NULL, NULL };

    if( test_atomic_function_set( deviceID, context, queue, num_elements, atom_and_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ false  ) != 0 )
        return -1;
    if( test_atomic_function_set( deviceID, context, queue, num_elements, atomic_and_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ true  ) != 0 )
        return -1;
    return 0;
}


#pragma mark ---- or

const char atom_or_core[] =
"    size_t numBits = sizeof( destMemory[0] ) * 8;\n"
"    int  whichResult = tid / numBits;\n"
"    int  bitIndex = tid - ( whichResult * numBits );\n"
"\n"
"    oldValues[tid] = atom_or( &destMemory[whichResult], ( 1L << bitIndex ) );\n"
;

const char atomic_or_core[] =
"    size_t numBits = sizeof( destMemory[0] ) * 8;\n"
"    int  whichResult = tid / numBits;\n"
"    int  bitIndex = tid - ( whichResult * numBits );\n"
"\n"
"    oldValues[tid] = atomic_or( &destMemory[whichResult], ( 1L << bitIndex ) );\n"
;

cl_int test_atomic_or_result_int( size_t size, cl_int *startRefValues, size_t whichResult )
{
    size_t numThreads = ( (size_t)size + 31 ) / 32;
    if( whichResult < numThreads - 1 )
        return 0xffffffff;

    // Last item doesn't get and'ed on every bit, so we have to mask away
    size_t numBits = (size_t)size - whichResult * 32;
    cl_int bits = 0;
    for( size_t i = 0; i < numBits; i++ )
        bits |= ( 1 << i );

    return bits;
}

cl_long test_atomic_or_result_long( size_t size, cl_long *startRefValues, size_t whichResult )
{
    size_t numThreads = ( (size_t)size + 63 ) / 64;
    if( whichResult < numThreads - 1 )
        return 0x0ffffffffffffffffLL;

    // Last item doesn't get and'ed on every bit, so we have to mask away
    size_t numBits = (size_t)size - whichResult * 64;
    cl_long bits = 0;
    for( size_t i = 0; i < numBits; i++ )
        bits |= ( 1LL << i );

    return bits;
}

int test_atomic_or(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { 0, 0LL, test_bitwise_num_results, test_atomic_or_result_int, NULL, NULL, test_atomic_or_result_long, NULL, NULL };

    if( test_atomic_function_set( deviceID, context, queue, num_elements, atom_or_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ false  ) != 0 )
        return -1;
    if( test_atomic_function_set( deviceID, context, queue, num_elements, atomic_or_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ true  ) != 0 )
        return -1;
    return 0;
}


#pragma mark ---- xor

const char atom_xor_core[] =
    "    size_t numBits = sizeof( destMemory[0] ) * 8;\n"
    "    int  bitIndex = tid & ( numBits - 1 );\n"
    "\n"
    "    oldValues[tid] = atom_xor( &destMemory[0], 1L << bitIndex );\n";

const char atomic_xor_core[] =
    "    size_t numBits = sizeof( destMemory[0] ) * 8;\n"
    "    int  bitIndex = tid & ( numBits - 1 );\n"
    "\n"
    "    oldValues[tid] = atomic_xor( &destMemory[0], 1L << bitIndex );\n";

cl_int test_atomic_xor_result_int( size_t size, cl_int *startRefValues, size_t whichResult )
{
    cl_int total = 0x2f08ab41;
    for( size_t i = 0; i < size; i++ )
        total ^= ( 1 << ( i & 31 ) );
    return total;
}

cl_long test_atomic_xor_result_long( size_t size, cl_long *startRefValues, size_t whichResult )
{
    cl_long total = 0x2f08ab418ba0541LL;
    for( size_t i = 0; i < size; i++ )
        total ^= ( 1LL << ( i & 63 ) );
    return total;
}

int test_atomic_xor(cl_device_id deviceID, cl_context context, cl_command_queue queue, int num_elements)
{
    TestFns set = { 0x2f08ab41, 0x2f08ab418ba0541LL, NULL, test_atomic_xor_result_int, NULL, NULL, test_atomic_xor_result_long, NULL, NULL };

    if( test_atomic_function_set( deviceID, context, queue, num_elements, atom_xor_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ false  ) != 0 )
        return -1;
    if( test_atomic_function_set( deviceID, context, queue, num_elements, atomic_xor_core, set, true, /*matchGroupSize*/ false, /*usingAtomicPrefix*/ true  ) != 0 )
        return -1;
    return 0;
}