🏡 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 "function_list.h"
#include "test_functions.h"
#include "utility.h"

#include <cstring>

const float twoToMinus126 = MAKE_HEX_FLOAT(0x1p-126f, 1, -126);

static int BuildKernel(const char *name, int vectorSize, cl_uint kernel_count,
                       cl_kernel *k, cl_program *p, bool relaxedMode)
{
    const char *c[] = { "__kernel void math_kernel",
                        sizeNames[vectorSize],
                        "( __global float",
                        sizeNames[vectorSize],
                        "* out, __global float",
                        sizeNames[vectorSize],
                        "* in1, __global float",
                        sizeNames[vectorSize],
                        "* in2 )\n"
                        "{\n"
                        "   size_t i = get_global_id(0);\n"
                        "   out[i] = ",
                        name,
                        "( in1[i], in2[i] );\n"
                        "}\n" };

    const char *c3[] = {
        "__kernel void math_kernel",
        sizeNames[vectorSize],
        "( __global float* out, __global float* in, __global float* in2)\n"
        "{\n"
        "   size_t i = get_global_id(0);\n"
        "   if( i + 1 < get_global_size(0) )\n"
        "   {\n"
        "       float3 f0 = vload3( 0, in + 3 * i );\n"
        "       float3 f1 = vload3( 0, in2 + 3 * i );\n"
        "       f0 = ",
        name,
        "( f0, f1 );\n"
        "       vstore3( f0, 0, out + 3*i );\n"
        "   }\n"
        "   else\n"
        "   {\n"
        "       size_t parity = i & 1;   // Figure out how many elements are "
        "left over after BUFFER_SIZE % (3*sizeof(float)). Assume power of two "
        "buffer size \n"
        "       float3 f0;\n"
        "       float3 f1;\n"
        "       switch( parity )\n"
        "       {\n"
        "           case 1:\n"
        "               f0 = (float3)( in[3*i], NAN, NAN ); \n"
        "               f1 = (float3)( in2[3*i], NAN, NAN ); \n"
        "               break;\n"
        "           case 0:\n"
        "               f0 = (float3)( in[3*i], in[3*i+1], NAN ); \n"
        "               f1 = (float3)( in2[3*i], in2[3*i+1], NAN ); \n"
        "               break;\n"
        "       }\n"
        "       f0 = ",
        name,
        "( f0, f1 );\n"
        "       switch( parity )\n"
        "       {\n"
        "           case 0:\n"
        "               out[3*i+1] = f0.y; \n"
        "               // fall through\n"
        "           case 1:\n"
        "               out[3*i] = f0.x; \n"
        "               break;\n"
        "       }\n"
        "   }\n"
        "}\n"
    };

    const char **kern = c;
    size_t kernSize = sizeof(c) / sizeof(c[0]);

    if (sizeValues[vectorSize] == 3)
    {
        kern = c3;
        kernSize = sizeof(c3) / sizeof(c3[0]);
    }

    char testName[32];
    snprintf(testName, sizeof(testName) - 1, "math_kernel%s",
             sizeNames[vectorSize]);

    return MakeKernels(kern, (cl_uint)kernSize, testName, kernel_count, k, p,
                       relaxedMode);
}

typedef struct BuildKernelInfo
{
    cl_uint offset; // the first vector size to build
    cl_uint kernel_count;
    cl_kernel **kernels;
    cl_program *programs;
    const char *nameInCode;
    bool relaxedMode; // Whether to build with -cl-fast-relaxed-math.
} BuildKernelInfo;

static cl_int BuildKernelFn(cl_uint job_id, cl_uint thread_id UNUSED, void *p)
{
    BuildKernelInfo *info = (BuildKernelInfo *)p;
    cl_uint i = info->offset + job_id;
    return BuildKernel(info->nameInCode, i, info->kernel_count,
                       info->kernels[i], info->programs + i, info->relaxedMode);
}

// Thread specific data for a worker thread
typedef struct ThreadInfo
{
    cl_mem inBuf; // input buffer for the thread
    cl_mem inBuf2; // input buffer for the thread
    cl_mem outBuf[VECTOR_SIZE_COUNT]; // output buffers for the thread
    float maxError; // max error value. Init to 0.
    double
        maxErrorValue; // position of the max error value (param 1).  Init to 0.
    double maxErrorValue2; // position of the max error value (param 2).  Init
                           // to 0.
    MTdata d;
    cl_command_queue tQueue; // per thread command queue to improve performance
} ThreadInfo;

typedef struct TestInfo
{
    size_t subBufferSize; // Size of the sub-buffer in elements
    const Func *f; // A pointer to the function info
    cl_program programs[VECTOR_SIZE_COUNT]; // programs for various vector sizes
    cl_kernel
        *k[VECTOR_SIZE_COUNT]; // arrays of thread-specific kernels for each
                               // worker thread:  k[vector_size][thread_id]
    ThreadInfo *
        tinfo; // An array of thread specific information for each worker thread
    cl_uint threadCount; // Number of worker threads
    cl_uint jobCount; // Number of jobs
    cl_uint step; // step between each chunk and the next.
    cl_uint scale; // stride between individual test values
    float ulps; // max_allowed ulps
    int ftz; // non-zero if running in flush to zero mode

    int isFDim;
    int skipNanInf;
    int isNextafter;
    bool relaxedMode; // True if test is running in relaxed mode, false
                      // otherwise.
} TestInfo;

// A table of more difficult cases to get right
static const float specialValues[] = {
    -NAN,
    -INFINITY,
    -FLT_MAX,
    MAKE_HEX_FLOAT(-0x1.000002p64f, -0x1000002L, 40),
    MAKE_HEX_FLOAT(-0x1.0p64f, -0x1L, 64),
    MAKE_HEX_FLOAT(-0x1.fffffep63f, -0x1fffffeL, 39),
    MAKE_HEX_FLOAT(-0x1.000002p63f, -0x1000002L, 39),
    MAKE_HEX_FLOAT(-0x1.0p63f, -0x1L, 63),
    MAKE_HEX_FLOAT(-0x1.fffffep62f, -0x1fffffeL, 38),
    MAKE_HEX_FLOAT(-0x1.000002p32f, -0x1000002L, 8),
    MAKE_HEX_FLOAT(-0x1.0p32f, -0x1L, 32),
    MAKE_HEX_FLOAT(-0x1.fffffep31f, -0x1fffffeL, 7),
    MAKE_HEX_FLOAT(-0x1.000002p31f, -0x1000002L, 7),
    MAKE_HEX_FLOAT(-0x1.0p31f, -0x1L, 31),
    MAKE_HEX_FLOAT(-0x1.fffffep30f, -0x1fffffeL, 6),
    -1000.f,
    -100.f,
    -4.0f,
    -3.5f,
    -3.0f,
    MAKE_HEX_FLOAT(-0x1.800002p1f, -0x1800002L, -23),
    -2.5f,
    MAKE_HEX_FLOAT(-0x1.7ffffep1f, -0x17ffffeL, -23),
    -2.0f,
    MAKE_HEX_FLOAT(-0x1.800002p0f, -0x1800002L, -24),
    -1.5f,
    MAKE_HEX_FLOAT(-0x1.7ffffep0f, -0x17ffffeL, -24),
    MAKE_HEX_FLOAT(-0x1.000002p0f, -0x1000002L, -24),
    -1.0f,
    MAKE_HEX_FLOAT(-0x1.fffffep-1f, -0x1fffffeL, -25),
    MAKE_HEX_FLOAT(-0x1.000002p-1f, -0x1000002L, -25),
    -0.5f,
    MAKE_HEX_FLOAT(-0x1.fffffep-2f, -0x1fffffeL, -26),
    MAKE_HEX_FLOAT(-0x1.000002p-2f, -0x1000002L, -26),
    -0.25f,
    MAKE_HEX_FLOAT(-0x1.fffffep-3f, -0x1fffffeL, -27),
    MAKE_HEX_FLOAT(-0x1.000002p-126f, -0x1000002L, -150),
    -FLT_MIN,
    MAKE_HEX_FLOAT(-0x0.fffffep-126f, -0x0fffffeL, -150),
    MAKE_HEX_FLOAT(-0x0.000ffep-126f, -0x0000ffeL, -150),
    MAKE_HEX_FLOAT(-0x0.0000fep-126f, -0x00000feL, -150),
    MAKE_HEX_FLOAT(-0x0.00000ep-126f, -0x000000eL, -150),
    MAKE_HEX_FLOAT(-0x0.00000cp-126f, -0x000000cL, -150),
    MAKE_HEX_FLOAT(-0x0.00000ap-126f, -0x000000aL, -150),
    MAKE_HEX_FLOAT(-0x0.000008p-126f, -0x0000008L, -150),
    MAKE_HEX_FLOAT(-0x0.000006p-126f, -0x0000006L, -150),
    MAKE_HEX_FLOAT(-0x0.000004p-126f, -0x0000004L, -150),
    MAKE_HEX_FLOAT(-0x0.000002p-126f, -0x0000002L, -150),
    -0.0f,

    +NAN,
    +INFINITY,
    +FLT_MAX,
    MAKE_HEX_FLOAT(+0x1.000002p64f, +0x1000002L, 40),
    MAKE_HEX_FLOAT(+0x1.0p64f, +0x1L, 64),
    MAKE_HEX_FLOAT(+0x1.fffffep63f, +0x1fffffeL, 39),
    MAKE_HEX_FLOAT(+0x1.000002p63f, +0x1000002L, 39),
    MAKE_HEX_FLOAT(+0x1.0p63f, +0x1L, 63),
    MAKE_HEX_FLOAT(+0x1.fffffep62f, +0x1fffffeL, 38),
    MAKE_HEX_FLOAT(+0x1.000002p32f, +0x1000002L, 8),
    MAKE_HEX_FLOAT(+0x1.0p32f, +0x1L, 32),
    MAKE_HEX_FLOAT(+0x1.fffffep31f, +0x1fffffeL, 7),
    MAKE_HEX_FLOAT(+0x1.000002p31f, +0x1000002L, 7),
    MAKE_HEX_FLOAT(+0x1.0p31f, +0x1L, 31),
    MAKE_HEX_FLOAT(+0x1.fffffep30f, +0x1fffffeL, 6),
    +1000.f,
    +100.f,
    +4.0f,
    +3.5f,
    +3.0f,
    MAKE_HEX_FLOAT(+0x1.800002p1f, +0x1800002L, -23),
    2.5f,
    MAKE_HEX_FLOAT(+0x1.7ffffep1f, +0x17ffffeL, -23),
    +2.0f,
    MAKE_HEX_FLOAT(+0x1.800002p0f, +0x1800002L, -24),
    1.5f,
    MAKE_HEX_FLOAT(+0x1.7ffffep0f, +0x17ffffeL, -24),
    MAKE_HEX_FLOAT(+0x1.000002p0f, +0x1000002L, -24),
    +1.0f,
    MAKE_HEX_FLOAT(+0x1.fffffep-1f, +0x1fffffeL, -25),
    MAKE_HEX_FLOAT(+0x1.000002p-1f, +0x1000002L, -25),
    +0.5f,
    MAKE_HEX_FLOAT(+0x1.fffffep-2f, +0x1fffffeL, -26),
    MAKE_HEX_FLOAT(+0x1.000002p-2f, +0x1000002L, -26),
    +0.25f,
    MAKE_HEX_FLOAT(+0x1.fffffep-3f, +0x1fffffeL, -27),
    MAKE_HEX_FLOAT(0x1.000002p-126f, 0x1000002L, -150),
    +FLT_MIN,
    MAKE_HEX_FLOAT(+0x0.fffffep-126f, +0x0fffffeL, -150),
    MAKE_HEX_FLOAT(+0x0.000ffep-126f, +0x0000ffeL, -150),
    MAKE_HEX_FLOAT(+0x0.0000fep-126f, +0x00000feL, -150),
    MAKE_HEX_FLOAT(+0x0.00000ep-126f, +0x000000eL, -150),
    MAKE_HEX_FLOAT(+0x0.00000cp-126f, +0x000000cL, -150),
    MAKE_HEX_FLOAT(+0x0.00000ap-126f, +0x000000aL, -150),
    MAKE_HEX_FLOAT(+0x0.000008p-126f, +0x0000008L, -150),
    MAKE_HEX_FLOAT(+0x0.000006p-126f, +0x0000006L, -150),
    MAKE_HEX_FLOAT(+0x0.000004p-126f, +0x0000004L, -150),
    MAKE_HEX_FLOAT(+0x0.000002p-126f, +0x0000002L, -150),
    +0.0f,
};

static const size_t specialValuesCount =
    sizeof(specialValues) / sizeof(specialValues[0]);

static cl_int Test(cl_uint job_id, cl_uint thread_id, void *data);

int TestFunc_Float_Float_Float(const Func *f, MTdata d, bool relaxedMode)
{
    TestInfo test_info;
    cl_int error;
    size_t i, j;
    float maxError = 0.0f;
    double maxErrorVal = 0.0;
    double maxErrorVal2 = 0.0;
    int skipTestingRelaxed = 0;

    logFunctionInfo(f->name, sizeof(cl_float), relaxedMode);

    // Init test_info
    memset(&test_info, 0, sizeof(test_info));
    test_info.threadCount = GetThreadCount();
    test_info.subBufferSize = BUFFER_SIZE
        / (sizeof(cl_float) * RoundUpToNextPowerOfTwo(test_info.threadCount));
    test_info.scale = getTestScale(sizeof(cl_float));

    if (gWimpyMode)
    {
        test_info.subBufferSize = gWimpyBufferSize
            / (sizeof(cl_float)
               * RoundUpToNextPowerOfTwo(test_info.threadCount));
    }

    test_info.step = (cl_uint)test_info.subBufferSize * test_info.scale;
    if (test_info.step / test_info.subBufferSize != test_info.scale)
    {
        // there was overflow
        test_info.jobCount = 1;
    }
    else
    {
        test_info.jobCount = (cl_uint)((1ULL << 32) / test_info.step);
    }

    test_info.f = f;
    test_info.ulps = gIsEmbedded ? f->float_embedded_ulps : f->float_ulps;
    test_info.ftz =
        f->ftz || gForceFTZ || 0 == (CL_FP_DENORM & gFloatCapabilities);
    test_info.relaxedMode = relaxedMode;
    test_info.isFDim = 0 == strcmp("fdim", f->nameInCode);
    test_info.skipNanInf = test_info.isFDim && !gInfNanSupport;
    test_info.isNextafter = 0 == strcmp("nextafter", f->nameInCode);

    // cl_kernels aren't thread safe, so we make one for each vector size for
    // every thread
    for (i = gMinVectorSizeIndex; i < gMaxVectorSizeIndex; i++)
    {
        size_t array_size = test_info.threadCount * sizeof(cl_kernel);
        test_info.k[i] = (cl_kernel *)malloc(array_size);
        if (NULL == test_info.k[i])
        {
            vlog_error("Error: Unable to allocate storage for kernels!\n");
            error = CL_OUT_OF_HOST_MEMORY;
            goto exit;
        }
        memset(test_info.k[i], 0, array_size);
    }
    test_info.tinfo =
        (ThreadInfo *)malloc(test_info.threadCount * sizeof(*test_info.tinfo));
    if (NULL == test_info.tinfo)
    {
        vlog_error(
            "Error: Unable to allocate storage for thread specific data.\n");
        error = CL_OUT_OF_HOST_MEMORY;
        goto exit;
    }
    memset(test_info.tinfo, 0,
           test_info.threadCount * sizeof(*test_info.tinfo));
    for (i = 0; i < test_info.threadCount; i++)
    {
        cl_buffer_region region = {
            i * test_info.subBufferSize * sizeof(cl_float),
            test_info.subBufferSize * sizeof(cl_float)
        };
        test_info.tinfo[i].inBuf =
            clCreateSubBuffer(gInBuffer, CL_MEM_READ_ONLY,
                              CL_BUFFER_CREATE_TYPE_REGION, &region, &error);
        if (error || NULL == test_info.tinfo[i].inBuf)
        {
            vlog_error("Error: Unable to create sub-buffer of gInBuffer for "
                       "region {%zd, %zd}\n",
                       region.origin, region.size);
            goto exit;
        }
        test_info.tinfo[i].inBuf2 =
            clCreateSubBuffer(gInBuffer2, CL_MEM_READ_ONLY,
                              CL_BUFFER_CREATE_TYPE_REGION, &region, &error);
        if (error || NULL == test_info.tinfo[i].inBuf2)
        {
            vlog_error("Error: Unable to create sub-buffer of gInBuffer2 for "
                       "region {%zd, %zd}\n",
                       region.origin, region.size);
            goto exit;
        }

        for (j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
        {
            test_info.tinfo[i].outBuf[j] = clCreateSubBuffer(
                gOutBuffer[j], CL_MEM_WRITE_ONLY, CL_BUFFER_CREATE_TYPE_REGION,
                &region, &error);
            if (error || NULL == test_info.tinfo[i].outBuf[j])
            {
                vlog_error("Error: Unable to create sub-buffer of "
                           "gOutBuffer[%d] for region {%zd, %zd}\n",
                           (int)j, region.origin, region.size);
                goto exit;
            }
        }
        test_info.tinfo[i].tQueue =
            clCreateCommandQueue(gContext, gDevice, 0, &error);
        if (NULL == test_info.tinfo[i].tQueue || error)
        {
            vlog_error("clCreateCommandQueue failed. (%d)\n", error);
            goto exit;
        }

        test_info.tinfo[i].d = init_genrand(genrand_int32(d));
    }

    // Init the kernels
    {
        BuildKernelInfo build_info = {
            gMinVectorSizeIndex, test_info.threadCount, test_info.k,
            test_info.programs,  f->nameInCode,         relaxedMode
        };
        if ((error = ThreadPool_Do(BuildKernelFn,
                                   gMaxVectorSizeIndex - gMinVectorSizeIndex,
                                   &build_info)))
            goto exit;
    }

    // Run the kernels
    if (!gSkipCorrectnessTesting)
    {
        error = ThreadPool_Do(Test, test_info.jobCount, &test_info);

        // Accumulate the arithmetic errors
        for (i = 0; i < test_info.threadCount; i++)
        {
            if (test_info.tinfo[i].maxError > maxError)
            {
                maxError = test_info.tinfo[i].maxError;
                maxErrorVal = test_info.tinfo[i].maxErrorValue;
                maxErrorVal2 = test_info.tinfo[i].maxErrorValue2;
            }
        }

        if (error) goto exit;

        if (gWimpyMode)
            vlog("Wimp pass");
        else
            vlog("passed");

        vlog("\t%8.2f @ {%a, %a}", maxError, maxErrorVal, maxErrorVal2);
    }

    vlog("\n");

exit:
    // Release
    for (i = gMinVectorSizeIndex; i < gMaxVectorSizeIndex; i++)
    {
        clReleaseProgram(test_info.programs[i]);
        if (test_info.k[i])
        {
            for (j = 0; j < test_info.threadCount; j++)
                clReleaseKernel(test_info.k[i][j]);

            free(test_info.k[i]);
        }
    }
    if (test_info.tinfo)
    {
        for (i = 0; i < test_info.threadCount; i++)
        {
            free_mtdata(test_info.tinfo[i].d);
            clReleaseMemObject(test_info.tinfo[i].inBuf);
            clReleaseMemObject(test_info.tinfo[i].inBuf2);
            for (j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
                clReleaseMemObject(test_info.tinfo[i].outBuf[j]);
            clReleaseCommandQueue(test_info.tinfo[i].tQueue);
        }

        free(test_info.tinfo);
    }

    return error;
}

static cl_int Test(cl_uint job_id, cl_uint thread_id, void *data)
{
    const TestInfo *job = (const TestInfo *)data;
    size_t buffer_elements = job->subBufferSize;
    size_t buffer_size = buffer_elements * sizeof(cl_float);
    cl_uint base = job_id * (cl_uint)job->step;
    ThreadInfo *tinfo = job->tinfo + thread_id;
    fptr func = job->f->func;
    int ftz = job->ftz;
    bool relaxedMode = job->relaxedMode;
    float ulps = getAllowedUlpError(job->f, relaxedMode);
    MTdata d = tinfo->d;
    cl_uint j, k;
    cl_int error;
    cl_uchar *overflow = (cl_uchar *)malloc(buffer_size);
    const char *name = job->f->name;
    int isFDim = job->isFDim;
    int skipNanInf = job->skipNanInf;
    int isNextafter = job->isNextafter;
    cl_uint *t = 0;
    cl_float *r = 0;
    cl_float *s = 0;
    cl_float *s2 = 0;
    cl_int copysign_test = 0;
    RoundingMode oldRoundMode;
    int skipVerification = 0;

    if (relaxedMode)
    {
        func = job->f->rfunc;
        if (strcmp(name, "pow") == 0 && gFastRelaxedDerived)
        {
            ulps = INFINITY;
            skipVerification = 1;
        }
    }

    // start the map of the output arrays
    cl_event e[VECTOR_SIZE_COUNT];
    cl_uint *out[VECTOR_SIZE_COUNT];
    for (j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
    {
        out[j] = (cl_uint *)clEnqueueMapBuffer(
            tinfo->tQueue, tinfo->outBuf[j], CL_FALSE, CL_MAP_WRITE, 0,
            buffer_size, 0, NULL, e + j, &error);
        if (error || NULL == out[j])
        {
            vlog_error("Error: clEnqueueMapBuffer %d failed! err: %d\n", j,
                       error);
            return error;
        }
    }

    // Get that moving
    if ((error = clFlush(tinfo->tQueue))) vlog("clFlush failed\n");

    // Init input array
    cl_uint *p = (cl_uint *)gIn + thread_id * buffer_elements;
    cl_uint *p2 = (cl_uint *)gIn2 + thread_id * buffer_elements;
    j = 0;

    int totalSpecialValueCount = specialValuesCount * specialValuesCount;
    int indx = (totalSpecialValueCount - 1) / buffer_elements;

    if (job_id <= (cl_uint)indx)
    { // test edge cases
        float *fp = (float *)p;
        float *fp2 = (float *)p2;
        uint32_t x, y;

        x = (job_id * buffer_elements) % specialValuesCount;
        y = (job_id * buffer_elements) / specialValuesCount;

        for (; j < buffer_elements; j++)
        {
            fp[j] = specialValues[x];
            fp2[j] = specialValues[y];
            ++x;
            if (x >= specialValuesCount)
            {
                x = 0;
                y++;
                if (y >= specialValuesCount) break;
            }
        }
    }

    // Init any remaining values.
    for (; j < buffer_elements; j++)
    {
        p[j] = genrand_int32(d);
        p2[j] = genrand_int32(d);
    }

    if ((error = clEnqueueWriteBuffer(tinfo->tQueue, tinfo->inBuf, CL_FALSE, 0,
                                      buffer_size, p, 0, NULL, NULL)))
    {
        vlog_error("Error: clEnqueueWriteBuffer failed! err: %d\n", error);
        goto exit;
    }

    if ((error = clEnqueueWriteBuffer(tinfo->tQueue, tinfo->inBuf2, CL_FALSE, 0,
                                      buffer_size, p2, 0, NULL, NULL)))
    {
        vlog_error("Error: clEnqueueWriteBuffer failed! err: %d\n", error);
        goto exit;
    }

    for (j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
    {
        // Wait for the map to finish
        if ((error = clWaitForEvents(1, e + j)))
        {
            vlog_error("Error: clWaitForEvents failed! err: %d\n", error);
            goto exit;
        }
        if ((error = clReleaseEvent(e[j])))
        {
            vlog_error("Error: clReleaseEvent failed! err: %d\n", error);
            goto exit;
        }

        // Fill the result buffer with garbage, so that old results don't carry
        // over
        uint32_t pattern = 0xffffdead;
        memset_pattern4(out[j], &pattern, buffer_size);
        if ((error = clEnqueueUnmapMemObject(tinfo->tQueue, tinfo->outBuf[j],
                                             out[j], 0, NULL, NULL)))
        {
            vlog_error("Error: clEnqueueMapBuffer failed! err: %d\n", error);
            goto exit;
        }

        // run the kernel
        size_t vectorCount =
            (buffer_elements + sizeValues[j] - 1) / sizeValues[j];
        cl_kernel kernel = job->k[j][thread_id]; // each worker thread has its
                                                 // own copy of the cl_kernel
        cl_program program = job->programs[j];

        if ((error = clSetKernelArg(kernel, 0, sizeof(tinfo->outBuf[j]),
                                    &tinfo->outBuf[j])))
        {
            LogBuildError(program);
            return error;
        }
        if ((error = clSetKernelArg(kernel, 1, sizeof(tinfo->inBuf),
                                    &tinfo->inBuf)))
        {
            LogBuildError(program);
            return error;
        }
        if ((error = clSetKernelArg(kernel, 2, sizeof(tinfo->inBuf2),
                                    &tinfo->inBuf2)))
        {
            LogBuildError(program);
            return error;
        }

        if ((error = clEnqueueNDRangeKernel(tinfo->tQueue, kernel, 1, NULL,
                                            &vectorCount, NULL, 0, NULL, NULL)))
        {
            vlog_error("FAILED -- could not execute kernel\n");
            goto exit;
        }
    }

    // Get that moving
    if ((error = clFlush(tinfo->tQueue))) vlog("clFlush 2 failed\n");

    if (gSkipCorrectnessTesting)
    {
        if ((error = clFinish(tinfo->tQueue)))
        {
            vlog_error("Error: clFinish failed! err: %d\n", error);
            goto exit;
        }
        free(overflow);
        return CL_SUCCESS;
    }

    FPU_mode_type oldMode;
    oldRoundMode = kRoundToNearestEven;
    if (isFDim)
    {
        // Calculate the correctly rounded reference result
        memset(&oldMode, 0, sizeof(oldMode));
        if (ftz) ForceFTZ(&oldMode);

        // Set the rounding mode to match the device
        if (gIsInRTZMode) oldRoundMode = set_round(kRoundTowardZero, kfloat);
    }

    if (!strcmp(name, "copysign")) copysign_test = 1;

#define ref_func(s, s2) (copysign_test ? func.f_ff_f(s, s2) : func.f_ff(s, s2))

    // Calculate the correctly rounded reference result
    r = (float *)gOut_Ref + thread_id * buffer_elements;
    s = (float *)gIn + thread_id * buffer_elements;
    s2 = (float *)gIn2 + thread_id * buffer_elements;
    if (skipNanInf)
    {
        for (j = 0; j < buffer_elements; j++)
        {
            feclearexcept(FE_OVERFLOW);
            r[j] = (float)ref_func(s[j], s2[j]);
            overflow[j] =
                FE_OVERFLOW == (FE_OVERFLOW & fetestexcept(FE_OVERFLOW));
        }
    }
    else
    {
        for (j = 0; j < buffer_elements; j++)
            r[j] = (float)ref_func(s[j], s2[j]);
    }

    if (isFDim && ftz) RestoreFPState(&oldMode);

    // Read the data back -- no need to wait for the first N-1 buffers. This is
    // an in order queue.
    for (j = gMinVectorSizeIndex; j + 1 < gMaxVectorSizeIndex; j++)
    {
        out[j] = (cl_uint *)clEnqueueMapBuffer(
            tinfo->tQueue, tinfo->outBuf[j], CL_FALSE, CL_MAP_READ, 0,
            buffer_size, 0, NULL, NULL, &error);
        if (error || NULL == out[j])
        {
            vlog_error("Error: clEnqueueMapBuffer %d failed! err: %d\n", j,
                       error);
            goto exit;
        }
    }

    // Wait for the last buffer
    out[j] = (cl_uint *)clEnqueueMapBuffer(tinfo->tQueue, tinfo->outBuf[j],
                                           CL_TRUE, CL_MAP_READ, 0, buffer_size,
                                           0, NULL, NULL, &error);
    if (error || NULL == out[j])
    {
        vlog_error("Error: clEnqueueMapBuffer %d failed! err: %d\n", j, error);
        goto exit;
    }

    if (!skipVerification)
    {
        // Verify data
        t = (cl_uint *)r;
        for (j = 0; j < buffer_elements; j++)
        {
            for (k = gMinVectorSizeIndex; k < gMaxVectorSizeIndex; k++)
            {
                cl_uint *q = out[k];

                // If we aren't getting the correctly rounded result
                if (t[j] != q[j])
                {
                    float test = ((float *)q)[j];
                    double correct = ref_func(s[j], s2[j]);

                    // Per section 10 paragraph 6, accept any result if an input
                    // or output is a infinity or NaN or overflow As per
                    // OpenCL 2.0 spec, section 5.8.4.3, enabling
                    // fast-relaxed-math mode also enables -cl-finite-math-only
                    // optimization. This optimization allows to assume that
                    // arguments and results are not NaNs or +/-INFs. Hence,
                    // accept any result if inputs or results are NaNs or INFs.
                    if (relaxedMode || skipNanInf)
                    {
                        if (skipNanInf && overflow[j]) continue;
                        // Note: no double rounding here.  Reference functions
                        // calculate in single precision.
                        if (IsFloatInfinity(correct) || IsFloatNaN(correct)
                            || IsFloatInfinity(s2[j]) || IsFloatNaN(s2[j])
                            || IsFloatInfinity(s[j]) || IsFloatNaN(s[j]))
                            continue;
                    }

                    float err = Ulp_Error(test, correct);
                    int fail = !(fabsf(err) <= ulps);

                    if (fail && ftz)
                    {
                        // retry per section 6.5.3.2
                        if (IsFloatResultSubnormal(correct, ulps))
                        {
                            fail = fail && (test != 0.0f);
                            if (!fail) err = 0.0f;
                        }

                        // nextafter on FTZ platforms may return the smallest
                        // normal float (2^-126) given a denormal or a zero
                        // as the first argument. The rationale here is that
                        // nextafter flushes the argument to zero and then
                        // returns the next representable number in the
                        // direction of the second argument, and since
                        // denorms are considered as zero, the smallest
                        // normal number is the next representable number.
                        // In which case, it should have the same sign as the
                        // second argument.
                        if (isNextafter)
                        {
                            if (IsFloatSubnormal(s[j]) || s[j] == 0.0f)
                            {
                                float value = copysignf(twoToMinus126, s2[j]);
                                fail = fail && (test != value);
                                if (!fail) err = 0.0f;
                            }
                        }
                        else
                        {
                            // retry per section 6.5.3.3
                            if (IsFloatSubnormal(s[j]))
                            {
                                double correct2, correct3;
                                float err2, err3;

                                if (skipNanInf) feclearexcept(FE_OVERFLOW);

                                correct2 = ref_func(0.0, s2[j]);
                                correct3 = ref_func(-0.0, s2[j]);

                                // Per section 10 paragraph 6, accept any result
                                // if an input or output is a infinity or NaN or
                                // overflow As per OpenCL 2.0 spec,
                                // section 5.8.4.3, enabling fast-relaxed-math
                                // mode also enables -cl-finite-math-only
                                // optimization. This optimization allows to
                                // assume that arguments and results are not
                                // NaNs or +/-INFs. Hence, accept any result if
                                // inputs or results are NaNs or INFs.
                                if (relaxedMode || skipNanInf)
                                {
                                    if (fetestexcept(FE_OVERFLOW) && skipNanInf)
                                        continue;

                                    // Note: no double rounding here.  Reference
                                    // functions calculate in single precision.
                                    if (IsFloatInfinity(correct2)
                                        || IsFloatNaN(correct2)
                                        || IsFloatInfinity(correct3)
                                        || IsFloatNaN(correct3))
                                        continue;
                                }

                                err2 = Ulp_Error(test, correct2);
                                err3 = Ulp_Error(test, correct3);
                                fail = fail
                                    && ((!(fabsf(err2) <= ulps))
                                        && (!(fabsf(err3) <= ulps)));
                                if (fabsf(err2) < fabsf(err)) err = err2;
                                if (fabsf(err3) < fabsf(err)) err = err3;

                                // retry per section 6.5.3.4
                                if (IsFloatResultSubnormal(correct2, ulps)
                                    || IsFloatResultSubnormal(correct3, ulps))
                                {
                                    fail = fail && (test != 0.0f);
                                    if (!fail) err = 0.0f;
                                }

                                // try with both args as zero
                                if (IsFloatSubnormal(s2[j]))
                                {
                                    double correct4, correct5;
                                    float err4, err5;

                                    if (skipNanInf) feclearexcept(FE_OVERFLOW);

                                    correct2 = ref_func(0.0, 0.0);
                                    correct3 = ref_func(-0.0, 0.0);
                                    correct4 = ref_func(0.0, -0.0);
                                    correct5 = ref_func(-0.0, -0.0);

                                    // Per section 10 paragraph 6, accept any
                                    // result if an input or output is a
                                    // infinity or NaN or overflow As per
                                    // OpenCL 2.0 spec, section 5.8.4.3,
                                    // enabling fast-relaxed-math mode also
                                    // enables -cl-finite-math-only
                                    // optimization. This optimization allows to
                                    // assume that arguments and results are not
                                    // NaNs or +/-INFs. Hence, accept any result
                                    // if inputs or results are NaNs or INFs.
                                    if (relaxedMode || skipNanInf)
                                    {
                                        if (fetestexcept(FE_OVERFLOW)
                                            && skipNanInf)
                                            continue;

                                        // Note: no double rounding here.
                                        // Reference functions calculate in
                                        // single precision.
                                        if (IsFloatInfinity(correct2)
                                            || IsFloatNaN(correct2)
                                            || IsFloatInfinity(correct3)
                                            || IsFloatNaN(correct3)
                                            || IsFloatInfinity(correct4)
                                            || IsFloatNaN(correct4)
                                            || IsFloatInfinity(correct5)
                                            || IsFloatNaN(correct5))
                                            continue;
                                    }

                                    err2 = Ulp_Error(test, correct2);
                                    err3 = Ulp_Error(test, correct3);
                                    err4 = Ulp_Error(test, correct4);
                                    err5 = Ulp_Error(test, correct5);
                                    fail = fail
                                        && ((!(fabsf(err2) <= ulps))
                                            && (!(fabsf(err3) <= ulps))
                                            && (!(fabsf(err4) <= ulps))
                                            && (!(fabsf(err5) <= ulps)));
                                    if (fabsf(err2) < fabsf(err)) err = err2;
                                    if (fabsf(err3) < fabsf(err)) err = err3;
                                    if (fabsf(err4) < fabsf(err)) err = err4;
                                    if (fabsf(err5) < fabsf(err)) err = err5;

                                    // retry per section 6.5.3.4
                                    if (IsFloatResultSubnormal(correct2, ulps)
                                        || IsFloatResultSubnormal(correct3,
                                                                  ulps)
                                        || IsFloatResultSubnormal(correct4,
                                                                  ulps)
                                        || IsFloatResultSubnormal(correct5,
                                                                  ulps))
                                    {
                                        fail = fail && (test != 0.0f);
                                        if (!fail) err = 0.0f;
                                    }
                                }
                            }
                            else if (IsFloatSubnormal(s2[j]))
                            {
                                double correct2, correct3;
                                float err2, err3;

                                if (skipNanInf) feclearexcept(FE_OVERFLOW);

                                correct2 = ref_func(s[j], 0.0);
                                correct3 = ref_func(s[j], -0.0);

                                // Per section 10 paragraph 6, accept any result
                                // if an input or output is a infinity or NaN or
                                // overflow As per OpenCL 2.0 spec,
                                // section 5.8.4.3, enabling fast-relaxed-math
                                // mode also enables -cl-finite-math-only
                                // optimization. This optimization allows to
                                // assume that arguments and results are not
                                // NaNs or +/-INFs. Hence, accept any result if
                                // inputs or results are NaNs or INFs.
                                if (relaxedMode || skipNanInf)
                                {
                                    // Note: no double rounding here.  Reference
                                    // functions calculate in single precision.
                                    if (overflow[j] && skipNanInf) continue;

                                    if (IsFloatInfinity(correct2)
                                        || IsFloatNaN(correct2)
                                        || IsFloatInfinity(correct3)
                                        || IsFloatNaN(correct3))
                                        continue;
                                }

                                err2 = Ulp_Error(test, correct2);
                                err3 = Ulp_Error(test, correct3);
                                fail = fail
                                    && ((!(fabsf(err2) <= ulps))
                                        && (!(fabsf(err3) <= ulps)));
                                if (fabsf(err2) < fabsf(err)) err = err2;
                                if (fabsf(err3) < fabsf(err)) err = err3;

                                // retry per section 6.5.3.4
                                if (IsFloatResultSubnormal(correct2, ulps)
                                    || IsFloatResultSubnormal(correct3, ulps))
                                {
                                    fail = fail && (test != 0.0f);
                                    if (!fail) err = 0.0f;
                                }
                            }
                        }
                    }

                    if (fabsf(err) > tinfo->maxError)
                    {
                        tinfo->maxError = fabsf(err);
                        tinfo->maxErrorValue = s[j];
                        tinfo->maxErrorValue2 = s2[j];
                    }
                    if (fail)
                    {
                        vlog_error(
                            "\nERROR: %s%s: %f ulp error at {%a (0x%x), %a "
                            "(0x%x)}: *%a vs. %a (0x%8.8x) at index: %d\n",
                            name, sizeNames[k], err, s[j], ((cl_uint *)s)[j],
                            s2[j], ((cl_uint *)s2)[j], r[j], test,
                            ((cl_uint *)&test)[0], j);
                        error = -1;
                        goto exit;
                    }
                }
            }
        }
    }

    if (isFDim && gIsInRTZMode) (void)set_round(oldRoundMode, kfloat);

    for (j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++)
    {
        if ((error = clEnqueueUnmapMemObject(tinfo->tQueue, tinfo->outBuf[j],
                                             out[j], 0, NULL, NULL)))
        {
            vlog_error("Error: clEnqueueUnmapMemObject %d failed 2! err: %d\n",
                       j, error);
            return error;
        }
    }

    if ((error = clFlush(tinfo->tQueue))) vlog("clFlush 3 failed\n");


    if (0 == (base & 0x0fffffff))
    {
        if (gVerboseBruteForce)
        {
            vlog("base:%14u step:%10u scale:%10zu buf_elements:%10u ulps:%5.3f "
                 "ThreadCount:%2u\n",
                 base, job->step, job->scale, buffer_elements, job->ulps,
                 job->threadCount);
        }
        else
        {
            vlog(".");
        }
        fflush(stdout);
    }

exit:
    if (overflow) free(overflow);
    return error;
}