mirror of
https://github.com/bkaradzic/bgfx.git
synced 2026-02-17 20:52:36 +01:00
Updated spirv-tools.
This commit is contained in:
Бранимир Караџић
@@ -1 +1 @@
|
||||
"v2022.5-dev", "SPIRV-Tools v2022.5-dev 466c6a4e95f6c2ebd61040d6fe2f6418e45b7e20"
|
||||
"v2023.2-dev", "SPIRV-Tools v2023.2-dev v2022.4-89-g4ee97c03"
|
||||
|
||||
@@ -95,7 +95,6 @@ static const spvtools::Extension pygen_variable_exts_SPV_KHR_ray_query[] = {spvt
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_KHR_ray_tracing[] = {spvtools::Extension::kSPV_KHR_ray_tracing};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_KHR_ray_tracingSPV_KHR_ray_query[] = {spvtools::Extension::kSPV_KHR_ray_tracing, spvtools::Extension::kSPV_KHR_ray_query};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_KHR_shader_ballot[] = {spvtools::Extension::kSPV_KHR_shader_ballot};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_KHR_shader_clock[] = {spvtools::Extension::kSPV_KHR_shader_clock};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_KHR_subgroup_vote[] = {spvtools::Extension::kSPV_KHR_subgroup_vote};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_KHR_terminate_invocation[] = {spvtools::Extension::kSPV_KHR_terminate_invocation};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_NV_cooperative_matrix[] = {spvtools::Extension::kSPV_NV_cooperative_matrix};
|
||||
@@ -494,7 +493,7 @@ static const spv_opcode_desc_t kOpcodeTableEntries[] = {
|
||||
{"GroupSMaxNonUniformAMD", spv::Op::OpGroupSMaxNonUniformAMD, 1, pygen_variable_caps_Groups, 5, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_SCOPE_ID, SPV_OPERAND_TYPE_GROUP_OPERATION, SPV_OPERAND_TYPE_ID}, 1, 1, 1, pygen_variable_exts_SPV_AMD_shader_ballot, 0xffffffffu, 0xffffffffu},
|
||||
{"FragmentMaskFetchAMD", spv::Op::OpFragmentMaskFetchAMD, 1, pygen_variable_caps_FragmentMaskAMD, 4, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 1, 1, 1, pygen_variable_exts_SPV_AMD_shader_fragment_mask, 0xffffffffu, 0xffffffffu},
|
||||
{"FragmentFetchAMD", spv::Op::OpFragmentFetchAMD, 1, pygen_variable_caps_FragmentMaskAMD, 5, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 1, 1, 1, pygen_variable_exts_SPV_AMD_shader_fragment_mask, 0xffffffffu, 0xffffffffu},
|
||||
{"ReadClockKHR", spv::Op::OpReadClockKHR, 1, pygen_variable_caps_ShaderClockKHR, 3, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_SCOPE_ID}, 1, 1, 1, pygen_variable_exts_SPV_KHR_shader_clock, 0xffffffffu, 0xffffffffu},
|
||||
{"ReadClockKHR", spv::Op::OpReadClockKHR, 1, pygen_variable_caps_ShaderClockKHR, 3, {SPV_OPERAND_TYPE_TYPE_ID, SPV_OPERAND_TYPE_RESULT_ID, SPV_OPERAND_TYPE_SCOPE_ID}, 1, 1, 0, nullptr, 0xffffffffu, 0xffffffffu},
|
||||
{"HitObjectRecordHitMotionNV", spv::Op::OpHitObjectRecordHitMotionNV, 2, pygen_variable_caps_ShaderInvocationReorderNVRayTracingMotionBlurNV, 14, {SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 0, 0, 0, nullptr, 0xffffffffu, 0xffffffffu},
|
||||
{"HitObjectRecordHitWithIndexMotionNV", spv::Op::OpHitObjectRecordHitWithIndexMotionNV, 2, pygen_variable_caps_ShaderInvocationReorderNVRayTracingMotionBlurNV, 13, {SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 0, 0, 0, nullptr, 0xffffffffu, 0xffffffffu},
|
||||
{"HitObjectRecordMissMotionNV", spv::Op::OpHitObjectRecordMissMotionNV, 2, pygen_variable_caps_ShaderInvocationReorderNVRayTracingMotionBlurNV, 7, {SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID, SPV_OPERAND_TYPE_ID}, 0, 0, 0, nullptr, 0xffffffffu, 0xffffffffu},
|
||||
|
||||
@@ -74,6 +74,8 @@ const char* ExtensionToString(Extension extension) {
|
||||
return "SPV_INTEL_float_controls2";
|
||||
case Extension::kSPV_INTEL_fp_fast_math_mode:
|
||||
return "SPV_INTEL_fp_fast_math_mode";
|
||||
case Extension::kSPV_INTEL_fpga_argument_interfaces:
|
||||
return "SPV_INTEL_fpga_argument_interfaces";
|
||||
case Extension::kSPV_INTEL_fpga_buffer_location:
|
||||
return "SPV_INTEL_fpga_buffer_location";
|
||||
case Extension::kSPV_INTEL_fpga_cluster_attributes:
|
||||
@@ -229,8 +231,8 @@ const char* ExtensionToString(Extension extension) {
|
||||
|
||||
|
||||
bool GetExtensionFromString(const char* str, Extension* extension) {
|
||||
static const char* known_ext_strs[] = { "SPV_AMD_gcn_shader", "SPV_AMD_gpu_shader_half_float", "SPV_AMD_gpu_shader_half_float_fetch", "SPV_AMD_gpu_shader_int16", "SPV_AMD_shader_ballot", "SPV_AMD_shader_early_and_late_fragment_tests", "SPV_AMD_shader_explicit_vertex_parameter", "SPV_AMD_shader_fragment_mask", "SPV_AMD_shader_image_load_store_lod", "SPV_AMD_shader_trinary_minmax", "SPV_AMD_texture_gather_bias_lod", "SPV_ARM_core_builtins", "SPV_EXT_demote_to_helper_invocation", "SPV_EXT_descriptor_indexing", "SPV_EXT_fragment_fully_covered", "SPV_EXT_fragment_invocation_density", "SPV_EXT_fragment_shader_interlock", "SPV_EXT_mesh_shader", "SPV_EXT_opacity_micromap", "SPV_EXT_physical_storage_buffer", "SPV_EXT_shader_atomic_float16_add", "SPV_EXT_shader_atomic_float_add", "SPV_EXT_shader_atomic_float_min_max", "SPV_EXT_shader_image_int64", "SPV_EXT_shader_stencil_export", "SPV_EXT_shader_viewport_index_layer", "SPV_GOOGLE_decorate_string", "SPV_GOOGLE_hlsl_functionality1", "SPV_GOOGLE_user_type", "SPV_INTEL_arbitrary_precision_fixed_point", "SPV_INTEL_arbitrary_precision_floating_point", "SPV_INTEL_arbitrary_precision_integers", "SPV_INTEL_blocking_pipes", "SPV_INTEL_debug_module", "SPV_INTEL_device_side_avc_motion_estimation", "SPV_INTEL_float_controls2", "SPV_INTEL_fp_fast_math_mode", "SPV_INTEL_fpga_buffer_location", "SPV_INTEL_fpga_cluster_attributes", "SPV_INTEL_fpga_dsp_control", "SPV_INTEL_fpga_invocation_pipelining_attributes", "SPV_INTEL_fpga_loop_controls", "SPV_INTEL_fpga_memory_accesses", "SPV_INTEL_fpga_memory_attributes", "SPV_INTEL_fpga_reg", "SPV_INTEL_function_pointers", "SPV_INTEL_inline_assembly", "SPV_INTEL_io_pipes", "SPV_INTEL_kernel_attributes", "SPV_INTEL_long_constant_composite", "SPV_INTEL_loop_fuse", "SPV_INTEL_media_block_io", "SPV_INTEL_memory_access_aliasing", "SPV_INTEL_optnone", "SPV_INTEL_runtime_aligned", "SPV_INTEL_shader_integer_functions2", "SPV_INTEL_split_barrier", "SPV_INTEL_subgroups", "SPV_INTEL_unstructured_loop_controls", "SPV_INTEL_usm_storage_classes", "SPV_INTEL_variable_length_array", "SPV_INTEL_vector_compute", "SPV_KHR_16bit_storage", "SPV_KHR_8bit_storage", "SPV_KHR_bit_instructions", "SPV_KHR_device_group", "SPV_KHR_expect_assume", "SPV_KHR_float_controls", "SPV_KHR_fragment_shader_barycentric", "SPV_KHR_fragment_shading_rate", "SPV_KHR_integer_dot_product", "SPV_KHR_linkonce_odr", "SPV_KHR_multiview", "SPV_KHR_no_integer_wrap_decoration", "SPV_KHR_non_semantic_info", "SPV_KHR_physical_storage_buffer", "SPV_KHR_post_depth_coverage", "SPV_KHR_ray_cull_mask", "SPV_KHR_ray_query", "SPV_KHR_ray_tracing", "SPV_KHR_shader_atomic_counter_ops", "SPV_KHR_shader_ballot", "SPV_KHR_shader_clock", "SPV_KHR_shader_draw_parameters", "SPV_KHR_storage_buffer_storage_class", "SPV_KHR_subgroup_rotate", "SPV_KHR_subgroup_uniform_control_flow", "SPV_KHR_subgroup_vote", "SPV_KHR_terminate_invocation", "SPV_KHR_uniform_group_instructions", "SPV_KHR_variable_pointers", "SPV_KHR_vulkan_memory_model", "SPV_KHR_workgroup_memory_explicit_layout", "SPV_NVX_multiview_per_view_attributes", "SPV_NV_bindless_texture", "SPV_NV_compute_shader_derivatives", "SPV_NV_cooperative_matrix", "SPV_NV_fragment_shader_barycentric", "SPV_NV_geometry_shader_passthrough", "SPV_NV_mesh_shader", "SPV_NV_ray_tracing", "SPV_NV_ray_tracing_motion_blur", "SPV_NV_sample_mask_override_coverage", "SPV_NV_shader_image_footprint", "SPV_NV_shader_invocation_reorder", "SPV_NV_shader_sm_builtins", "SPV_NV_shader_subgroup_partitioned", "SPV_NV_shading_rate", "SPV_NV_stereo_view_rendering", "SPV_NV_viewport_array2", "SPV_VALIDATOR_ignore_type_decl_unique" };
|
||||
static const Extension known_ext_ids[] = { Extension::kSPV_AMD_gcn_shader, Extension::kSPV_AMD_gpu_shader_half_float, Extension::kSPV_AMD_gpu_shader_half_float_fetch, Extension::kSPV_AMD_gpu_shader_int16, Extension::kSPV_AMD_shader_ballot, Extension::kSPV_AMD_shader_early_and_late_fragment_tests, Extension::kSPV_AMD_shader_explicit_vertex_parameter, Extension::kSPV_AMD_shader_fragment_mask, Extension::kSPV_AMD_shader_image_load_store_lod, Extension::kSPV_AMD_shader_trinary_minmax, Extension::kSPV_AMD_texture_gather_bias_lod, Extension::kSPV_ARM_core_builtins, Extension::kSPV_EXT_demote_to_helper_invocation, Extension::kSPV_EXT_descriptor_indexing, Extension::kSPV_EXT_fragment_fully_covered, Extension::kSPV_EXT_fragment_invocation_density, Extension::kSPV_EXT_fragment_shader_interlock, Extension::kSPV_EXT_mesh_shader, Extension::kSPV_EXT_opacity_micromap, Extension::kSPV_EXT_physical_storage_buffer, Extension::kSPV_EXT_shader_atomic_float16_add, Extension::kSPV_EXT_shader_atomic_float_add, Extension::kSPV_EXT_shader_atomic_float_min_max, Extension::kSPV_EXT_shader_image_int64, Extension::kSPV_EXT_shader_stencil_export, Extension::kSPV_EXT_shader_viewport_index_layer, Extension::kSPV_GOOGLE_decorate_string, Extension::kSPV_GOOGLE_hlsl_functionality1, Extension::kSPV_GOOGLE_user_type, Extension::kSPV_INTEL_arbitrary_precision_fixed_point, Extension::kSPV_INTEL_arbitrary_precision_floating_point, Extension::kSPV_INTEL_arbitrary_precision_integers, Extension::kSPV_INTEL_blocking_pipes, Extension::kSPV_INTEL_debug_module, Extension::kSPV_INTEL_device_side_avc_motion_estimation, Extension::kSPV_INTEL_float_controls2, Extension::kSPV_INTEL_fp_fast_math_mode, Extension::kSPV_INTEL_fpga_buffer_location, Extension::kSPV_INTEL_fpga_cluster_attributes, Extension::kSPV_INTEL_fpga_dsp_control, Extension::kSPV_INTEL_fpga_invocation_pipelining_attributes, Extension::kSPV_INTEL_fpga_loop_controls, Extension::kSPV_INTEL_fpga_memory_accesses, Extension::kSPV_INTEL_fpga_memory_attributes, Extension::kSPV_INTEL_fpga_reg, Extension::kSPV_INTEL_function_pointers, Extension::kSPV_INTEL_inline_assembly, Extension::kSPV_INTEL_io_pipes, Extension::kSPV_INTEL_kernel_attributes, Extension::kSPV_INTEL_long_constant_composite, Extension::kSPV_INTEL_loop_fuse, Extension::kSPV_INTEL_media_block_io, Extension::kSPV_INTEL_memory_access_aliasing, Extension::kSPV_INTEL_optnone, Extension::kSPV_INTEL_runtime_aligned, Extension::kSPV_INTEL_shader_integer_functions2, Extension::kSPV_INTEL_split_barrier, Extension::kSPV_INTEL_subgroups, Extension::kSPV_INTEL_unstructured_loop_controls, Extension::kSPV_INTEL_usm_storage_classes, Extension::kSPV_INTEL_variable_length_array, Extension::kSPV_INTEL_vector_compute, Extension::kSPV_KHR_16bit_storage, Extension::kSPV_KHR_8bit_storage, Extension::kSPV_KHR_bit_instructions, Extension::kSPV_KHR_device_group, Extension::kSPV_KHR_expect_assume, Extension::kSPV_KHR_float_controls, Extension::kSPV_KHR_fragment_shader_barycentric, Extension::kSPV_KHR_fragment_shading_rate, Extension::kSPV_KHR_integer_dot_product, Extension::kSPV_KHR_linkonce_odr, Extension::kSPV_KHR_multiview, Extension::kSPV_KHR_no_integer_wrap_decoration, Extension::kSPV_KHR_non_semantic_info, Extension::kSPV_KHR_physical_storage_buffer, Extension::kSPV_KHR_post_depth_coverage, Extension::kSPV_KHR_ray_cull_mask, Extension::kSPV_KHR_ray_query, Extension::kSPV_KHR_ray_tracing, Extension::kSPV_KHR_shader_atomic_counter_ops, Extension::kSPV_KHR_shader_ballot, Extension::kSPV_KHR_shader_clock, Extension::kSPV_KHR_shader_draw_parameters, Extension::kSPV_KHR_storage_buffer_storage_class, Extension::kSPV_KHR_subgroup_rotate, Extension::kSPV_KHR_subgroup_uniform_control_flow, Extension::kSPV_KHR_subgroup_vote, Extension::kSPV_KHR_terminate_invocation, Extension::kSPV_KHR_uniform_group_instructions, Extension::kSPV_KHR_variable_pointers, Extension::kSPV_KHR_vulkan_memory_model, Extension::kSPV_KHR_workgroup_memory_explicit_layout, Extension::kSPV_NVX_multiview_per_view_attributes, Extension::kSPV_NV_bindless_texture, Extension::kSPV_NV_compute_shader_derivatives, Extension::kSPV_NV_cooperative_matrix, Extension::kSPV_NV_fragment_shader_barycentric, Extension::kSPV_NV_geometry_shader_passthrough, Extension::kSPV_NV_mesh_shader, Extension::kSPV_NV_ray_tracing, Extension::kSPV_NV_ray_tracing_motion_blur, Extension::kSPV_NV_sample_mask_override_coverage, Extension::kSPV_NV_shader_image_footprint, Extension::kSPV_NV_shader_invocation_reorder, Extension::kSPV_NV_shader_sm_builtins, Extension::kSPV_NV_shader_subgroup_partitioned, Extension::kSPV_NV_shading_rate, Extension::kSPV_NV_stereo_view_rendering, Extension::kSPV_NV_viewport_array2, Extension::kSPV_VALIDATOR_ignore_type_decl_unique };
|
||||
static const char* known_ext_strs[] = { "SPV_AMD_gcn_shader", "SPV_AMD_gpu_shader_half_float", "SPV_AMD_gpu_shader_half_float_fetch", "SPV_AMD_gpu_shader_int16", "SPV_AMD_shader_ballot", "SPV_AMD_shader_early_and_late_fragment_tests", "SPV_AMD_shader_explicit_vertex_parameter", "SPV_AMD_shader_fragment_mask", "SPV_AMD_shader_image_load_store_lod", "SPV_AMD_shader_trinary_minmax", "SPV_AMD_texture_gather_bias_lod", "SPV_ARM_core_builtins", "SPV_EXT_demote_to_helper_invocation", "SPV_EXT_descriptor_indexing", "SPV_EXT_fragment_fully_covered", "SPV_EXT_fragment_invocation_density", "SPV_EXT_fragment_shader_interlock", "SPV_EXT_mesh_shader", "SPV_EXT_opacity_micromap", "SPV_EXT_physical_storage_buffer", "SPV_EXT_shader_atomic_float16_add", "SPV_EXT_shader_atomic_float_add", "SPV_EXT_shader_atomic_float_min_max", "SPV_EXT_shader_image_int64", "SPV_EXT_shader_stencil_export", "SPV_EXT_shader_viewport_index_layer", "SPV_GOOGLE_decorate_string", "SPV_GOOGLE_hlsl_functionality1", "SPV_GOOGLE_user_type", "SPV_INTEL_arbitrary_precision_fixed_point", "SPV_INTEL_arbitrary_precision_floating_point", "SPV_INTEL_arbitrary_precision_integers", "SPV_INTEL_blocking_pipes", "SPV_INTEL_debug_module", "SPV_INTEL_device_side_avc_motion_estimation", "SPV_INTEL_float_controls2", "SPV_INTEL_fp_fast_math_mode", "SPV_INTEL_fpga_argument_interfaces", "SPV_INTEL_fpga_buffer_location", "SPV_INTEL_fpga_cluster_attributes", "SPV_INTEL_fpga_dsp_control", "SPV_INTEL_fpga_invocation_pipelining_attributes", "SPV_INTEL_fpga_loop_controls", "SPV_INTEL_fpga_memory_accesses", "SPV_INTEL_fpga_memory_attributes", "SPV_INTEL_fpga_reg", "SPV_INTEL_function_pointers", "SPV_INTEL_inline_assembly", "SPV_INTEL_io_pipes", "SPV_INTEL_kernel_attributes", "SPV_INTEL_long_constant_composite", "SPV_INTEL_loop_fuse", "SPV_INTEL_media_block_io", "SPV_INTEL_memory_access_aliasing", "SPV_INTEL_optnone", "SPV_INTEL_runtime_aligned", "SPV_INTEL_shader_integer_functions2", "SPV_INTEL_split_barrier", "SPV_INTEL_subgroups", "SPV_INTEL_unstructured_loop_controls", "SPV_INTEL_usm_storage_classes", "SPV_INTEL_variable_length_array", "SPV_INTEL_vector_compute", "SPV_KHR_16bit_storage", "SPV_KHR_8bit_storage", "SPV_KHR_bit_instructions", "SPV_KHR_device_group", "SPV_KHR_expect_assume", "SPV_KHR_float_controls", "SPV_KHR_fragment_shader_barycentric", "SPV_KHR_fragment_shading_rate", "SPV_KHR_integer_dot_product", "SPV_KHR_linkonce_odr", "SPV_KHR_multiview", "SPV_KHR_no_integer_wrap_decoration", "SPV_KHR_non_semantic_info", "SPV_KHR_physical_storage_buffer", "SPV_KHR_post_depth_coverage", "SPV_KHR_ray_cull_mask", "SPV_KHR_ray_query", "SPV_KHR_ray_tracing", "SPV_KHR_shader_atomic_counter_ops", "SPV_KHR_shader_ballot", "SPV_KHR_shader_clock", "SPV_KHR_shader_draw_parameters", "SPV_KHR_storage_buffer_storage_class", "SPV_KHR_subgroup_rotate", "SPV_KHR_subgroup_uniform_control_flow", "SPV_KHR_subgroup_vote", "SPV_KHR_terminate_invocation", "SPV_KHR_uniform_group_instructions", "SPV_KHR_variable_pointers", "SPV_KHR_vulkan_memory_model", "SPV_KHR_workgroup_memory_explicit_layout", "SPV_NVX_multiview_per_view_attributes", "SPV_NV_bindless_texture", "SPV_NV_compute_shader_derivatives", "SPV_NV_cooperative_matrix", "SPV_NV_fragment_shader_barycentric", "SPV_NV_geometry_shader_passthrough", "SPV_NV_mesh_shader", "SPV_NV_ray_tracing", "SPV_NV_ray_tracing_motion_blur", "SPV_NV_sample_mask_override_coverage", "SPV_NV_shader_image_footprint", "SPV_NV_shader_invocation_reorder", "SPV_NV_shader_sm_builtins", "SPV_NV_shader_subgroup_partitioned", "SPV_NV_shading_rate", "SPV_NV_stereo_view_rendering", "SPV_NV_viewport_array2", "SPV_VALIDATOR_ignore_type_decl_unique" };
|
||||
static const Extension known_ext_ids[] = { Extension::kSPV_AMD_gcn_shader, Extension::kSPV_AMD_gpu_shader_half_float, Extension::kSPV_AMD_gpu_shader_half_float_fetch, Extension::kSPV_AMD_gpu_shader_int16, Extension::kSPV_AMD_shader_ballot, Extension::kSPV_AMD_shader_early_and_late_fragment_tests, Extension::kSPV_AMD_shader_explicit_vertex_parameter, Extension::kSPV_AMD_shader_fragment_mask, Extension::kSPV_AMD_shader_image_load_store_lod, Extension::kSPV_AMD_shader_trinary_minmax, Extension::kSPV_AMD_texture_gather_bias_lod, Extension::kSPV_ARM_core_builtins, Extension::kSPV_EXT_demote_to_helper_invocation, Extension::kSPV_EXT_descriptor_indexing, Extension::kSPV_EXT_fragment_fully_covered, Extension::kSPV_EXT_fragment_invocation_density, Extension::kSPV_EXT_fragment_shader_interlock, Extension::kSPV_EXT_mesh_shader, Extension::kSPV_EXT_opacity_micromap, Extension::kSPV_EXT_physical_storage_buffer, Extension::kSPV_EXT_shader_atomic_float16_add, Extension::kSPV_EXT_shader_atomic_float_add, Extension::kSPV_EXT_shader_atomic_float_min_max, Extension::kSPV_EXT_shader_image_int64, Extension::kSPV_EXT_shader_stencil_export, Extension::kSPV_EXT_shader_viewport_index_layer, Extension::kSPV_GOOGLE_decorate_string, Extension::kSPV_GOOGLE_hlsl_functionality1, Extension::kSPV_GOOGLE_user_type, Extension::kSPV_INTEL_arbitrary_precision_fixed_point, Extension::kSPV_INTEL_arbitrary_precision_floating_point, Extension::kSPV_INTEL_arbitrary_precision_integers, Extension::kSPV_INTEL_blocking_pipes, Extension::kSPV_INTEL_debug_module, Extension::kSPV_INTEL_device_side_avc_motion_estimation, Extension::kSPV_INTEL_float_controls2, Extension::kSPV_INTEL_fp_fast_math_mode, Extension::kSPV_INTEL_fpga_argument_interfaces, Extension::kSPV_INTEL_fpga_buffer_location, Extension::kSPV_INTEL_fpga_cluster_attributes, Extension::kSPV_INTEL_fpga_dsp_control, Extension::kSPV_INTEL_fpga_invocation_pipelining_attributes, Extension::kSPV_INTEL_fpga_loop_controls, Extension::kSPV_INTEL_fpga_memory_accesses, Extension::kSPV_INTEL_fpga_memory_attributes, Extension::kSPV_INTEL_fpga_reg, Extension::kSPV_INTEL_function_pointers, Extension::kSPV_INTEL_inline_assembly, Extension::kSPV_INTEL_io_pipes, Extension::kSPV_INTEL_kernel_attributes, Extension::kSPV_INTEL_long_constant_composite, Extension::kSPV_INTEL_loop_fuse, Extension::kSPV_INTEL_media_block_io, Extension::kSPV_INTEL_memory_access_aliasing, Extension::kSPV_INTEL_optnone, Extension::kSPV_INTEL_runtime_aligned, Extension::kSPV_INTEL_shader_integer_functions2, Extension::kSPV_INTEL_split_barrier, Extension::kSPV_INTEL_subgroups, Extension::kSPV_INTEL_unstructured_loop_controls, Extension::kSPV_INTEL_usm_storage_classes, Extension::kSPV_INTEL_variable_length_array, Extension::kSPV_INTEL_vector_compute, Extension::kSPV_KHR_16bit_storage, Extension::kSPV_KHR_8bit_storage, Extension::kSPV_KHR_bit_instructions, Extension::kSPV_KHR_device_group, Extension::kSPV_KHR_expect_assume, Extension::kSPV_KHR_float_controls, Extension::kSPV_KHR_fragment_shader_barycentric, Extension::kSPV_KHR_fragment_shading_rate, Extension::kSPV_KHR_integer_dot_product, Extension::kSPV_KHR_linkonce_odr, Extension::kSPV_KHR_multiview, Extension::kSPV_KHR_no_integer_wrap_decoration, Extension::kSPV_KHR_non_semantic_info, Extension::kSPV_KHR_physical_storage_buffer, Extension::kSPV_KHR_post_depth_coverage, Extension::kSPV_KHR_ray_cull_mask, Extension::kSPV_KHR_ray_query, Extension::kSPV_KHR_ray_tracing, Extension::kSPV_KHR_shader_atomic_counter_ops, Extension::kSPV_KHR_shader_ballot, Extension::kSPV_KHR_shader_clock, Extension::kSPV_KHR_shader_draw_parameters, Extension::kSPV_KHR_storage_buffer_storage_class, Extension::kSPV_KHR_subgroup_rotate, Extension::kSPV_KHR_subgroup_uniform_control_flow, Extension::kSPV_KHR_subgroup_vote, Extension::kSPV_KHR_terminate_invocation, Extension::kSPV_KHR_uniform_group_instructions, Extension::kSPV_KHR_variable_pointers, Extension::kSPV_KHR_vulkan_memory_model, Extension::kSPV_KHR_workgroup_memory_explicit_layout, Extension::kSPV_NVX_multiview_per_view_attributes, Extension::kSPV_NV_bindless_texture, Extension::kSPV_NV_compute_shader_derivatives, Extension::kSPV_NV_cooperative_matrix, Extension::kSPV_NV_fragment_shader_barycentric, Extension::kSPV_NV_geometry_shader_passthrough, Extension::kSPV_NV_mesh_shader, Extension::kSPV_NV_ray_tracing, Extension::kSPV_NV_ray_tracing_motion_blur, Extension::kSPV_NV_sample_mask_override_coverage, Extension::kSPV_NV_shader_image_footprint, Extension::kSPV_NV_shader_invocation_reorder, Extension::kSPV_NV_shader_sm_builtins, Extension::kSPV_NV_shader_subgroup_partitioned, Extension::kSPV_NV_shading_rate, Extension::kSPV_NV_stereo_view_rendering, Extension::kSPV_NV_viewport_array2, Extension::kSPV_VALIDATOR_ignore_type_decl_unique };
|
||||
const auto b = std::begin(known_ext_strs);
|
||||
const auto e = std::end(known_ext_strs);
|
||||
const auto found = std::equal_range(
|
||||
@@ -656,6 +658,8 @@ const char* CapabilityToString(spv::Capability capability) {
|
||||
return "DebugInfoModuleINTEL";
|
||||
case spv::Capability::SplitBarrierINTEL:
|
||||
return "SplitBarrierINTEL";
|
||||
case spv::Capability::FPGAArgumentInterfacesINTEL:
|
||||
return "FPGAArgumentInterfacesINTEL";
|
||||
case spv::Capability::GroupUniformArithmeticKHR:
|
||||
return "GroupUniformArithmeticKHR";
|
||||
case spv::Capability::Max:
|
||||
|
||||
@@ -35,6 +35,7 @@ kSPV_INTEL_debug_module,
|
||||
kSPV_INTEL_device_side_avc_motion_estimation,
|
||||
kSPV_INTEL_float_controls2,
|
||||
kSPV_INTEL_fp_fast_math_mode,
|
||||
kSPV_INTEL_fpga_argument_interfaces,
|
||||
kSPV_INTEL_fpga_buffer_location,
|
||||
kSPV_INTEL_fpga_cluster_attributes,
|
||||
kSPV_INTEL_fpga_dsp_control,
|
||||
|
||||
@@ -15,6 +15,7 @@ static const spv::Capability pygen_variable_caps_DeviceGroup[] = {spv::Capabilit
|
||||
static const spv::Capability pygen_variable_caps_DrawParameters[] = {spv::Capability::DrawParameters};
|
||||
static const spv::Capability pygen_variable_caps_DrawParametersMeshShadingNVMeshShadingEXT[] = {spv::Capability::DrawParameters, spv::Capability::MeshShadingNV, spv::Capability::MeshShadingEXT};
|
||||
static const spv::Capability pygen_variable_caps_FPFastMathModeINTEL[] = {spv::Capability::FPFastMathModeINTEL};
|
||||
static const spv::Capability pygen_variable_caps_FPGAArgumentInterfacesINTEL[] = {spv::Capability::FPGAArgumentInterfacesINTEL};
|
||||
static const spv::Capability pygen_variable_caps_FPGABufferLocationINTEL[] = {spv::Capability::FPGABufferLocationINTEL};
|
||||
static const spv::Capability pygen_variable_caps_FPGAClusterAttributesINTEL[] = {spv::Capability::FPGAClusterAttributesINTEL};
|
||||
static const spv::Capability pygen_variable_caps_FPGADSPControlINTEL[] = {spv::Capability::FPGADSPControlINTEL};
|
||||
@@ -164,6 +165,7 @@ static const spvtools::Extension pygen_variable_exts_SPV_INTEL_debug_module[] =
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_INTEL_device_side_avc_motion_estimation[] = {spvtools::Extension::kSPV_INTEL_device_side_avc_motion_estimation};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_INTEL_float_controls2[] = {spvtools::Extension::kSPV_INTEL_float_controls2};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_INTEL_fp_fast_math_mode[] = {spvtools::Extension::kSPV_INTEL_fp_fast_math_mode};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_INTEL_fpga_argument_interfaces[] = {spvtools::Extension::kSPV_INTEL_fpga_argument_interfaces};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_INTEL_fpga_buffer_location[] = {spvtools::Extension::kSPV_INTEL_fpga_buffer_location};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_INTEL_fpga_cluster_attributes[] = {spvtools::Extension::kSPV_INTEL_fpga_cluster_attributes};
|
||||
static const spvtools::Extension pygen_variable_exts_SPV_INTEL_fpga_dsp_control[] = {spvtools::Extension::kSPV_INTEL_fpga_dsp_control};
|
||||
@@ -831,7 +833,16 @@ static const spv_operand_desc_t pygen_variable_DecorationEntries[] = {
|
||||
{"FunctionFloatingPointModeINTEL", 6080, 1, pygen_variable_caps_FunctionFloatControlINTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER, SPV_OPERAND_TYPE_FPOPERATION_MODE}, 0xffffffffu, 0xffffffffu},
|
||||
{"SingleElementVectorINTEL", 6085, 1, pygen_variable_caps_VectorComputeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"VectorComputeCallableFunctionINTEL", 6087, 1, pygen_variable_caps_VectorComputeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"MediaBlockIOINTEL", 6140, 1, pygen_variable_caps_VectorComputeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu}
|
||||
{"MediaBlockIOINTEL", 6140, 1, pygen_variable_caps_VectorComputeINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"ConduitKernelArgumentINTEL", 6175, 1, pygen_variable_caps_FPGAArgumentInterfacesINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"RegisterMapKernelArgumentINTEL", 6176, 1, pygen_variable_caps_FPGAArgumentInterfacesINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"MMHostInterfaceAddressWidthINTEL", 6177, 1, pygen_variable_caps_FPGAArgumentInterfacesINTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu},
|
||||
{"MMHostInterfaceDataWidthINTEL", 6178, 1, pygen_variable_caps_FPGAArgumentInterfacesINTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu},
|
||||
{"MMHostInterfaceLatencyINTEL", 6179, 1, pygen_variable_caps_FPGAArgumentInterfacesINTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu},
|
||||
{"MMHostInterfaceReadWriteModeINTEL", 6180, 1, pygen_variable_caps_FPGAArgumentInterfacesINTEL, 0, nullptr, {SPV_OPERAND_TYPE_ACCESS_QUALIFIER}, 0xffffffffu, 0xffffffffu},
|
||||
{"MMHostInterfaceMaxBurstINTEL", 6181, 1, pygen_variable_caps_FPGAArgumentInterfacesINTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu},
|
||||
{"MMHostInterfaceWaitRequestINTEL", 6182, 1, pygen_variable_caps_FPGAArgumentInterfacesINTEL, 0, nullptr, {SPV_OPERAND_TYPE_LITERAL_INTEGER}, 0xffffffffu, 0xffffffffu},
|
||||
{"StableKernelArgumentINTEL", 6183, 1, pygen_variable_caps_FPGAArgumentInterfacesINTEL, 0, nullptr, {}, 0xffffffffu, 0xffffffffu}
|
||||
};
|
||||
|
||||
static const spv_operand_desc_t pygen_variable_BuiltInEntries[] = {
|
||||
@@ -1104,7 +1115,7 @@ static const spv_operand_desc_t pygen_variable_CapabilityEntries[] = {
|
||||
{"StencilExportEXT", 5013, 1, pygen_variable_caps_Shader, 1, pygen_variable_exts_SPV_EXT_shader_stencil_export, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"ImageReadWriteLodAMD", 5015, 1, pygen_variable_caps_Shader, 1, pygen_variable_exts_SPV_AMD_shader_image_load_store_lod, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"Int64ImageEXT", 5016, 1, pygen_variable_caps_Shader, 1, pygen_variable_exts_SPV_EXT_shader_image_int64, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"ShaderClockKHR", 5055, 1, pygen_variable_caps_Shader, 1, pygen_variable_exts_SPV_KHR_shader_clock, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"ShaderClockKHR", 5055, 0, nullptr, 1, pygen_variable_exts_SPV_KHR_shader_clock, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"SampleMaskOverrideCoverageNV", 5249, 1, pygen_variable_caps_SampleRateShading, 1, pygen_variable_exts_SPV_NV_sample_mask_override_coverage, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"GeometryShaderPassthroughNV", 5251, 1, pygen_variable_caps_Geometry, 1, pygen_variable_exts_SPV_NV_geometry_shader_passthrough, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"ShaderViewportIndexLayerEXT", 5254, 1, pygen_variable_caps_MultiViewport, 2, pygen_variable_exts_SPV_EXT_shader_viewport_index_layerSPV_NV_viewport_array2, {}, 0xffffffffu, 0xffffffffu},
|
||||
@@ -1226,6 +1237,7 @@ static const spv_operand_desc_t pygen_variable_CapabilityEntries[] = {
|
||||
{"AtomicFloat16AddEXT", 6095, 0, nullptr, 1, pygen_variable_exts_SPV_EXT_shader_atomic_float16_add, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"DebugInfoModuleINTEL", 6114, 0, nullptr, 1, pygen_variable_exts_SPV_INTEL_debug_module, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"SplitBarrierINTEL", 6141, 0, nullptr, 1, pygen_variable_exts_SPV_INTEL_split_barrier, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"FPGAArgumentInterfacesINTEL", 6174, 0, nullptr, 1, pygen_variable_exts_SPV_INTEL_fpga_argument_interfaces, {}, 0xffffffffu, 0xffffffffu},
|
||||
{"GroupUniformArithmeticKHR", 6400, 0, nullptr, 1, pygen_variable_exts_SPV_KHR_uniform_group_instructions, {}, 0xffffffffu, 0xffffffffu}
|
||||
};
|
||||
|
||||
|
||||
@@ -442,6 +442,8 @@ typedef struct spv_reducer_options_t spv_reducer_options_t;
|
||||
|
||||
typedef struct spv_fuzzer_options_t spv_fuzzer_options_t;
|
||||
|
||||
typedef struct spv_optimizer_t spv_optimizer_t;
|
||||
|
||||
// Type Definitions
|
||||
|
||||
typedef spv_const_binary_t* spv_const_binary;
|
||||
@@ -901,6 +903,63 @@ SPIRV_TOOLS_EXPORT spv_result_t spvBinaryParse(
|
||||
const size_t num_words, spv_parsed_header_fn_t parse_header,
|
||||
spv_parsed_instruction_fn_t parse_instruction, spv_diagnostic* diagnostic);
|
||||
|
||||
// The optimizer interface.
|
||||
|
||||
// A pointer to a function that accepts a log message from an optimizer.
|
||||
typedef void (*spv_message_consumer)(
|
||||
spv_message_level_t, const char*, const spv_position_t*, const char*);
|
||||
|
||||
// Creates and returns an optimizer object. This object must be passed to
|
||||
// optimizer APIs below and is valid until passed to spvOptimizerDestroy.
|
||||
SPIRV_TOOLS_EXPORT spv_optimizer_t* spvOptimizerCreate(spv_target_env env);
|
||||
|
||||
// Destroys the given optimizer object.
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerDestroy(spv_optimizer_t* optimizer);
|
||||
|
||||
// Sets an spv_message_consumer on an optimizer object.
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerSetMessageConsumer(
|
||||
spv_optimizer_t* optimizer, spv_message_consumer consumer);
|
||||
|
||||
// Registers passes that attempt to legalize the generated code.
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerRegisterLegalizationPasses(
|
||||
spv_optimizer_t* optimizer);
|
||||
|
||||
// Registers passes that attempt to improve performance of generated code.
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerRegisterPerformancePasses(
|
||||
spv_optimizer_t* optimizer);
|
||||
|
||||
// Registers passes that attempt to improve the size of generated code.
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerRegisterSizePasses(
|
||||
spv_optimizer_t* optimizer);
|
||||
|
||||
// Registers a pass specified by a flag in an optimizer object.
|
||||
SPIRV_TOOLS_EXPORT bool spvOptimizerRegisterPassFromFlag(
|
||||
spv_optimizer_t* optimizer, const char* flag);
|
||||
|
||||
// Registers passes specified by length number of flags in an optimizer object.
|
||||
SPIRV_TOOLS_EXPORT bool spvOptimizerRegisterPassesFromFlags(
|
||||
spv_optimizer_t* optimizer, const char** flags, const size_t flag_count);
|
||||
|
||||
// Optimizes the SPIR-V code of size |word_count| pointed to by |binary| and
|
||||
// returns an optimized spv_binary in |optimized_binary|.
|
||||
//
|
||||
// Returns SPV_SUCCESS on successful optimization, whether or not the module is
|
||||
// modified. Returns an SPV_ERROR_* if the module fails to validate or if
|
||||
// errors occur when processing using any of the registered passes. In that
|
||||
// case, no further passes are executed and the |optimized_binary| contents may
|
||||
// be invalid.
|
||||
//
|
||||
// By default, the binary is validated before any transforms are performed,
|
||||
// and optionally after each transform. Validation uses SPIR-V spec rules
|
||||
// for the SPIR-V version named in the binary's header (at word offset 1).
|
||||
// Additionally, if the target environment is a client API (such as
|
||||
// Vulkan 1.1), then validate for that client API version, to the extent
|
||||
// that it is verifiable from data in the binary itself, or from the
|
||||
// validator options set on the optimizer options.
|
||||
SPIRV_TOOLS_EXPORT spv_result_t spvOptimizerRun(
|
||||
spv_optimizer_t* optimizer, const uint32_t* binary, const size_t word_count,
|
||||
spv_binary* optimized_binary, const spv_optimizer_options options);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
|
||||
@@ -497,8 +497,8 @@ uint32_t InstBindlessCheckPass::GenLastByteIdx(RefAnalysis* ref,
|
||||
if (sum_id == 0)
|
||||
sum_id = curr_offset_id;
|
||||
else {
|
||||
Instruction* sum_inst = builder->AddBinaryOp(GetUintId(), spv::Op::OpIAdd,
|
||||
sum_id, curr_offset_id);
|
||||
Instruction* sum_inst =
|
||||
builder->AddIAdd(GetUintId(), sum_id, curr_offset_id);
|
||||
sum_id = sum_inst->result_id();
|
||||
}
|
||||
++ac_in_idx;
|
||||
@@ -507,8 +507,7 @@ uint32_t InstBindlessCheckPass::GenLastByteIdx(RefAnalysis* ref,
|
||||
uint32_t bsize = ByteSize(curr_ty_id, matrix_stride, col_major, in_matrix);
|
||||
uint32_t last = bsize - 1;
|
||||
uint32_t last_id = builder->GetUintConstantId(last);
|
||||
Instruction* sum_inst =
|
||||
builder->AddBinaryOp(GetUintId(), spv::Op::OpIAdd, sum_id, last_id);
|
||||
Instruction* sum_inst = builder->AddIAdd(GetUintId(), sum_id, last_id);
|
||||
return sum_inst->result_id();
|
||||
}
|
||||
|
||||
@@ -536,6 +535,8 @@ void InstBindlessCheckPass::GenCheckCode(
|
||||
new BasicBlock(std::move(valid_label)));
|
||||
builder.SetInsertPoint(&*new_blk_ptr);
|
||||
uint32_t new_ref_id = CloneOriginalReference(ref, &builder);
|
||||
uint32_t null_id = 0;
|
||||
uint32_t ref_type_id = ref->ref_inst->type_id();
|
||||
(void)builder.AddBranch(merge_blk_id);
|
||||
new_blocks->push_back(std::move(new_blk_ptr));
|
||||
// Gen invalid block
|
||||
@@ -563,10 +564,23 @@ void InstBindlessCheckPass::GenCheckCode(
|
||||
GenDebugStreamWrite(uid2offset_[ref->ref_inst->unique_id()], stage_idx,
|
||||
{error_id, u_index_id, u_length_id}, &builder);
|
||||
}
|
||||
// Generate a ConstantNull, converting to uint64 if the type cannot be a null.
|
||||
if (new_ref_id != 0) {
|
||||
analysis::TypeManager* type_mgr = context()->get_type_mgr();
|
||||
analysis::Type* ref_type = type_mgr->GetType(ref_type_id);
|
||||
if (ref_type->AsPointer() != nullptr) {
|
||||
context()->AddCapability(spv::Capability::Int64);
|
||||
uint32_t null_u64_id = GetNullId(GetUint64Id());
|
||||
Instruction* null_ptr_inst = builder.AddUnaryOp(
|
||||
ref_type_id, spv::Op::OpConvertUToPtr, null_u64_id);
|
||||
null_id = null_ptr_inst->result_id();
|
||||
} else {
|
||||
null_id = GetNullId(ref_type_id);
|
||||
}
|
||||
}
|
||||
// Remember last invalid block id
|
||||
uint32_t last_invalid_blk_id = new_blk_ptr->GetLabelInst()->result_id();
|
||||
// Gen zero for invalid reference
|
||||
uint32_t ref_type_id = ref->ref_inst->type_id();
|
||||
(void)builder.AddBranch(merge_blk_id);
|
||||
new_blocks->push_back(std::move(new_blk_ptr));
|
||||
// Gen merge block
|
||||
@@ -577,8 +591,7 @@ void InstBindlessCheckPass::GenCheckCode(
|
||||
// reference.
|
||||
if (new_ref_id != 0) {
|
||||
Instruction* phi_inst = builder.AddPhi(
|
||||
ref_type_id, {new_ref_id, valid_blk_id, GetNullId(ref_type_id),
|
||||
last_invalid_blk_id});
|
||||
ref_type_id, {new_ref_id, valid_blk_id, null_id, last_invalid_blk_id});
|
||||
context()->ReplaceAllUsesWith(ref->ref_inst->result_id(),
|
||||
phi_inst->result_id());
|
||||
}
|
||||
@@ -734,15 +747,7 @@ void InstBindlessCheckPass::GenTexBuffCheckCode(
|
||||
if (image_ty_inst->GetSingleWordInOperand(kSpvTypeImageArrayed) != 0) return;
|
||||
if (image_ty_inst->GetSingleWordInOperand(kSpvTypeImageMS) != 0) return;
|
||||
// Enable ImageQuery Capability if not yet enabled
|
||||
if (!get_feature_mgr()->HasCapability(spv::Capability::ImageQuery)) {
|
||||
std::unique_ptr<Instruction> cap_image_query_inst(
|
||||
new Instruction(context(), spv::Op::OpCapability, 0, 0,
|
||||
std::initializer_list<Operand>{
|
||||
{SPV_OPERAND_TYPE_CAPABILITY,
|
||||
{uint32_t(spv::Capability::ImageQuery)}}}));
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*cap_image_query_inst);
|
||||
context()->AddCapability(std::move(cap_image_query_inst));
|
||||
}
|
||||
context()->AddCapability(spv::Capability::ImageQuery);
|
||||
// Move original block's preceding instructions into first new block
|
||||
std::unique_ptr<BasicBlock> new_blk_ptr;
|
||||
MovePreludeCode(ref_inst_itr, ref_block_itr, &new_blk_ptr);
|
||||
|
||||
@@ -257,9 +257,7 @@ uint32_t InstBuffAddrCheckPass::GetSearchAndTestFuncId() {
|
||||
uint32_t hdr_blk_id = TakeNextId();
|
||||
// Branch to search loop header
|
||||
std::unique_ptr<Instruction> hdr_blk_label(NewLabel(hdr_blk_id));
|
||||
(void)builder.AddInstruction(MakeUnique<Instruction>(
|
||||
context(), spv::Op::OpBranch, 0, 0,
|
||||
std::initializer_list<Operand>{{SPV_OPERAND_TYPE_ID, {hdr_blk_id}}}));
|
||||
(void)builder.AddBranch(hdr_blk_id);
|
||||
input_func->AddBasicBlock(std::move(first_blk_ptr));
|
||||
// Linear search loop header block
|
||||
// TODO(greg-lunarg): Implement binary search
|
||||
@@ -293,17 +291,10 @@ uint32_t InstBuffAddrCheckPass::GetSearchAndTestFuncId() {
|
||||
uint32_t bound_test_blk_id = TakeNextId();
|
||||
std::unique_ptr<Instruction> bound_test_blk_label(
|
||||
NewLabel(bound_test_blk_id));
|
||||
(void)builder.AddInstruction(MakeUnique<Instruction>(
|
||||
context(), spv::Op::OpLoopMerge, 0, 0,
|
||||
std::initializer_list<Operand>{
|
||||
{SPV_OPERAND_TYPE_ID, {bound_test_blk_id}},
|
||||
{SPV_OPERAND_TYPE_ID, {cont_blk_id}},
|
||||
{SPV_OPERAND_TYPE_LITERAL_INTEGER,
|
||||
{uint32_t(spv::LoopControlMask::MaskNone)}}}));
|
||||
(void)builder.AddLoopMerge(bound_test_blk_id, cont_blk_id,
|
||||
uint32_t(spv::LoopControlMask::MaskNone));
|
||||
// Branch to continue/work block
|
||||
(void)builder.AddInstruction(MakeUnique<Instruction>(
|
||||
context(), spv::Op::OpBranch, 0, 0,
|
||||
std::initializer_list<Operand>{{SPV_OPERAND_TYPE_ID, {cont_blk_id}}}));
|
||||
(void)builder.AddBranch(cont_blk_id);
|
||||
input_func->AddBasicBlock(std::move(hdr_blk_ptr));
|
||||
// Continue/Work Block. Read next buffer pointer and break if greater
|
||||
// than ref_ptr arg.
|
||||
@@ -386,10 +377,8 @@ uint32_t InstBuffAddrCheckPass::GetSearchAndTestFuncId() {
|
||||
GetBoolId(), spv::Op::OpULessThanEqual, ref_end_inst->result_id(),
|
||||
len_load_inst->result_id());
|
||||
// Return test result
|
||||
(void)builder.AddInstruction(MakeUnique<Instruction>(
|
||||
context(), spv::Op::OpReturnValue, 0, 0,
|
||||
std::initializer_list<Operand>{
|
||||
{SPV_OPERAND_TYPE_ID, {len_test_inst->result_id()}}}));
|
||||
(void)builder.AddUnaryOp(0, spv::Op::OpReturnValue,
|
||||
len_test_inst->result_id());
|
||||
// Close block
|
||||
input_func->AddBasicBlock(std::move(bound_test_blk_ptr));
|
||||
// Close function and add function to module
|
||||
@@ -408,14 +397,7 @@ uint32_t InstBuffAddrCheckPass::GenSearchAndTest(Instruction* ref_inst,
|
||||
InstructionBuilder* builder,
|
||||
uint32_t* ref_uptr_id) {
|
||||
// Enable Int64 if necessary
|
||||
if (!get_feature_mgr()->HasCapability(spv::Capability::Int64)) {
|
||||
std::unique_ptr<Instruction> cap_int64_inst(new Instruction(
|
||||
context(), spv::Op::OpCapability, 0, 0,
|
||||
std::initializer_list<Operand>{{SPV_OPERAND_TYPE_CAPABILITY,
|
||||
{uint32_t(spv::Capability::Int64)}}}));
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*cap_int64_inst);
|
||||
context()->AddCapability(std::move(cap_int64_inst));
|
||||
}
|
||||
context()->AddCapability(spv::Capability::Int64);
|
||||
// Convert reference pointer to uint64
|
||||
uint32_t ref_ptr_id = ref_inst->GetSingleWordInOperand(0);
|
||||
Instruction* ref_uptr_inst =
|
||||
@@ -429,10 +411,8 @@ uint32_t InstBuffAddrCheckPass::GenSearchAndTest(Instruction* ref_inst,
|
||||
uint32_t ref_len = GetTypeLength(ref_ptr_ty_inst->GetSingleWordInOperand(1));
|
||||
uint32_t ref_len_id = builder->GetUintConstantId(ref_len);
|
||||
// Gen call to search and test function
|
||||
const std::vector<uint32_t> args = {GetSearchAndTestFuncId(), *ref_uptr_id,
|
||||
ref_len_id};
|
||||
Instruction* call_inst =
|
||||
builder->AddNaryOp(GetBoolId(), spv::Op::OpFunctionCall, args);
|
||||
Instruction* call_inst = builder->AddFunctionCall(
|
||||
GetBoolId(), GetSearchAndTestFuncId(), {*ref_uptr_id, ref_len_id});
|
||||
uint32_t retval = call_inst->result_id();
|
||||
return retval;
|
||||
}
|
||||
|
||||
@@ -34,8 +34,8 @@ void InstDebugPrintfPass::GenOutputValues(Instruction* val_inst,
|
||||
const analysis::Type* c_ty = v_ty->element_type();
|
||||
uint32_t c_ty_id = type_mgr->GetId(c_ty);
|
||||
for (uint32_t c = 0; c < v_ty->element_count(); ++c) {
|
||||
Instruction* c_inst = builder->AddIdLiteralOp(
|
||||
c_ty_id, spv::Op::OpCompositeExtract, val_inst->result_id(), c);
|
||||
Instruction* c_inst =
|
||||
builder->AddCompositeExtract(c_ty_id, val_inst->result_id(), {c});
|
||||
GenOutputValues(c_inst, val_ids, builder);
|
||||
}
|
||||
return;
|
||||
@@ -44,9 +44,8 @@ void InstDebugPrintfPass::GenOutputValues(Instruction* val_inst,
|
||||
// Select between uint32 zero or one
|
||||
uint32_t zero_id = builder->GetUintConstantId(0);
|
||||
uint32_t one_id = builder->GetUintConstantId(1);
|
||||
Instruction* sel_inst =
|
||||
builder->AddTernaryOp(GetUintId(), spv::Op::OpSelect,
|
||||
val_inst->result_id(), one_id, zero_id);
|
||||
Instruction* sel_inst = builder->AddSelect(
|
||||
GetUintId(), val_inst->result_id(), one_id, zero_id);
|
||||
val_ids->push_back(sel_inst->result_id());
|
||||
return;
|
||||
}
|
||||
|
||||
357
3rdparty/spirv-tools/source/opt/instrument_pass.cpp
vendored
357
3rdparty/spirv-tools/source/opt/instrument_pass.cpp
vendored
@@ -54,7 +54,6 @@ void InstrumentPass::MovePreludeCode(
|
||||
|
||||
void InstrumentPass::MovePostludeCode(
|
||||
UptrVectorIterator<BasicBlock> ref_block_itr, BasicBlock* new_blk_ptr) {
|
||||
// new_blk_ptr->reset(new BasicBlock(NewLabel(ref_block_itr->id())));
|
||||
// Move contents of original ref block.
|
||||
for (auto cii = ref_block_itr->begin(); cii != ref_block_itr->end();
|
||||
cii = ref_block_itr->begin()) {
|
||||
@@ -77,21 +76,62 @@ void InstrumentPass::MovePostludeCode(
|
||||
}
|
||||
|
||||
std::unique_ptr<Instruction> InstrumentPass::NewLabel(uint32_t label_id) {
|
||||
std::unique_ptr<Instruction> newLabel(
|
||||
new Instruction(context(), spv::Op::OpLabel, 0, label_id, {}));
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*newLabel);
|
||||
return newLabel;
|
||||
auto new_label =
|
||||
MakeUnique<Instruction>(context(), spv::Op::OpLabel, 0, label_id,
|
||||
std::initializer_list<Operand>{});
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*new_label);
|
||||
return new_label;
|
||||
}
|
||||
|
||||
std::unique_ptr<Function> InstrumentPass::StartFunction(
|
||||
uint32_t func_id, const analysis::Type* return_type,
|
||||
const std::vector<const analysis::Type*>& param_types) {
|
||||
analysis::TypeManager* type_mgr = context()->get_type_mgr();
|
||||
analysis::Function* func_type = GetFunction(return_type, param_types);
|
||||
|
||||
const std::vector<Operand> operands{
|
||||
{spv_operand_type_t::SPV_OPERAND_TYPE_LITERAL_INTEGER,
|
||||
{uint32_t(spv::FunctionControlMask::MaskNone)}},
|
||||
{spv_operand_type_t::SPV_OPERAND_TYPE_ID, {type_mgr->GetId(func_type)}},
|
||||
};
|
||||
auto func_inst =
|
||||
MakeUnique<Instruction>(context(), spv::Op::OpFunction,
|
||||
type_mgr->GetId(return_type), func_id, operands);
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*func_inst);
|
||||
return MakeUnique<Function>(std::move(func_inst));
|
||||
}
|
||||
|
||||
std::unique_ptr<Instruction> InstrumentPass::EndFunction() {
|
||||
auto end = MakeUnique<Instruction>(context(), spv::Op::OpFunctionEnd, 0, 0,
|
||||
std::initializer_list<Operand>{});
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(end.get());
|
||||
return end;
|
||||
}
|
||||
|
||||
std::vector<uint32_t> InstrumentPass::AddParameters(
|
||||
Function& func, const std::vector<const analysis::Type*>& param_types) {
|
||||
std::vector<uint32_t> param_ids;
|
||||
param_ids.reserve(param_types.size());
|
||||
for (const analysis::Type* param : param_types) {
|
||||
uint32_t pid = TakeNextId();
|
||||
param_ids.push_back(pid);
|
||||
auto param_inst =
|
||||
MakeUnique<Instruction>(context(), spv::Op::OpFunctionParameter,
|
||||
context()->get_type_mgr()->GetId(param), pid,
|
||||
std::initializer_list<Operand>{});
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(param_inst.get());
|
||||
func.AddParameter(std::move(param_inst));
|
||||
}
|
||||
return param_ids;
|
||||
}
|
||||
|
||||
std::unique_ptr<Instruction> InstrumentPass::NewName(
|
||||
uint32_t id, const std::string& name_str) {
|
||||
std::unique_ptr<Instruction> new_name(new Instruction(
|
||||
return MakeUnique<Instruction>(
|
||||
context(), spv::Op::OpName, 0, 0,
|
||||
std::initializer_list<Operand>{
|
||||
{SPV_OPERAND_TYPE_ID, {id}},
|
||||
{SPV_OPERAND_TYPE_LITERAL_STRING, utils::MakeVector(name_str)}}));
|
||||
|
||||
return new_name;
|
||||
{SPV_OPERAND_TYPE_LITERAL_STRING, utils::MakeVector(name_str)}});
|
||||
}
|
||||
|
||||
std::unique_ptr<Instruction> InstrumentPass::NewGlobalName(
|
||||
@@ -118,14 +158,12 @@ std::unique_ptr<Instruction> InstrumentPass::NewGlobalName(
|
||||
|
||||
std::unique_ptr<Instruction> InstrumentPass::NewMemberName(
|
||||
uint32_t id, uint32_t member_index, const std::string& name_str) {
|
||||
std::unique_ptr<Instruction> new_name(new Instruction(
|
||||
return MakeUnique<Instruction>(
|
||||
context(), spv::Op::OpMemberName, 0, 0,
|
||||
std::initializer_list<Operand>{
|
||||
{SPV_OPERAND_TYPE_ID, {id}},
|
||||
{SPV_OPERAND_TYPE_LITERAL_INTEGER, {member_index}},
|
||||
{SPV_OPERAND_TYPE_LITERAL_STRING, utils::MakeVector(name_str)}}));
|
||||
|
||||
return new_name;
|
||||
{SPV_OPERAND_TYPE_LITERAL_STRING, utils::MakeVector(name_str)}});
|
||||
}
|
||||
|
||||
uint32_t InstrumentPass::Gen32BitCvtCode(uint32_t val_id,
|
||||
@@ -167,17 +205,15 @@ void InstrumentPass::GenDebugOutputFieldCode(uint32_t base_offset_id,
|
||||
// Cast value to 32-bit unsigned if necessary
|
||||
uint32_t val_id = GenUintCastCode(field_value_id, builder);
|
||||
// Store value
|
||||
Instruction* data_idx_inst =
|
||||
builder->AddBinaryOp(GetUintId(), spv::Op::OpIAdd, base_offset_id,
|
||||
builder->GetUintConstantId(field_offset));
|
||||
Instruction* data_idx_inst = builder->AddIAdd(
|
||||
GetUintId(), base_offset_id, builder->GetUintConstantId(field_offset));
|
||||
uint32_t buf_id = GetOutputBufferId();
|
||||
uint32_t buf_uint_ptr_id = GetOutputBufferPtrId();
|
||||
Instruction* achain_inst =
|
||||
builder->AddTernaryOp(buf_uint_ptr_id, spv::Op::OpAccessChain, buf_id,
|
||||
builder->GetUintConstantId(kDebugOutputDataOffset),
|
||||
data_idx_inst->result_id());
|
||||
(void)builder->AddBinaryOp(0, spv::Op::OpStore, achain_inst->result_id(),
|
||||
val_id);
|
||||
Instruction* achain_inst = builder->AddAccessChain(
|
||||
buf_uint_ptr_id, buf_id,
|
||||
{builder->GetUintConstantId(kDebugOutputDataOffset),
|
||||
data_idx_inst->result_id()});
|
||||
(void)builder->AddStore(achain_inst->result_id(), val_id);
|
||||
}
|
||||
|
||||
void InstrumentPass::GenCommonStreamWriteCode(uint32_t record_sz,
|
||||
@@ -202,8 +238,8 @@ void InstrumentPass::GenCommonStreamWriteCode(uint32_t record_sz,
|
||||
void InstrumentPass::GenFragCoordEltDebugOutputCode(
|
||||
uint32_t base_offset_id, uint32_t uint_frag_coord_id, uint32_t element,
|
||||
InstructionBuilder* builder) {
|
||||
Instruction* element_val_inst = builder->AddIdLiteralOp(
|
||||
GetUintId(), spv::Op::OpCompositeExtract, uint_frag_coord_id, element);
|
||||
Instruction* element_val_inst =
|
||||
builder->AddCompositeExtract(GetUintId(), uint_frag_coord_id, {element});
|
||||
GenDebugOutputFieldCode(base_offset_id, kInstFragOutFragCoordX + element,
|
||||
element_val_inst->result_id(), builder);
|
||||
}
|
||||
@@ -212,8 +248,7 @@ uint32_t InstrumentPass::GenVarLoad(uint32_t var_id,
|
||||
InstructionBuilder* builder) {
|
||||
Instruction* var_inst = get_def_use_mgr()->GetDef(var_id);
|
||||
uint32_t type_id = GetPointeeTypeId(var_inst);
|
||||
Instruction* load_inst =
|
||||
builder->AddUnaryOp(type_id, spv::Op::OpLoad, var_id);
|
||||
Instruction* load_inst = builder->AddLoad(type_id, var_id);
|
||||
return load_inst->result_id();
|
||||
}
|
||||
|
||||
@@ -249,12 +284,12 @@ void InstrumentPass::GenStageStreamWriteCode(uint32_t stage_idx,
|
||||
uint32_t load_id = GenVarLoad(context()->GetBuiltinInputVarId(uint32_t(
|
||||
spv::BuiltIn::GlobalInvocationId)),
|
||||
builder);
|
||||
Instruction* x_inst = builder->AddIdLiteralOp(
|
||||
GetUintId(), spv::Op::OpCompositeExtract, load_id, 0);
|
||||
Instruction* y_inst = builder->AddIdLiteralOp(
|
||||
GetUintId(), spv::Op::OpCompositeExtract, load_id, 1);
|
||||
Instruction* z_inst = builder->AddIdLiteralOp(
|
||||
GetUintId(), spv::Op::OpCompositeExtract, load_id, 2);
|
||||
Instruction* x_inst =
|
||||
builder->AddCompositeExtract(GetUintId(), load_id, {0});
|
||||
Instruction* y_inst =
|
||||
builder->AddCompositeExtract(GetUintId(), load_id, {1});
|
||||
Instruction* z_inst =
|
||||
builder->AddCompositeExtract(GetUintId(), load_id, {2});
|
||||
GenDebugOutputFieldCode(base_offset_id, kInstCompOutGlobalInvocationIdX,
|
||||
x_inst->result_id(), builder);
|
||||
GenDebugOutputFieldCode(base_offset_id, kInstCompOutGlobalInvocationIdY,
|
||||
@@ -291,10 +326,10 @@ void InstrumentPass::GenStageStreamWriteCode(uint32_t stage_idx,
|
||||
Instruction* uvec3_cast_inst =
|
||||
builder->AddUnaryOp(GetVec3UintId(), spv::Op::OpBitcast, load_id);
|
||||
uint32_t uvec3_cast_id = uvec3_cast_inst->result_id();
|
||||
Instruction* u_inst = builder->AddIdLiteralOp(
|
||||
GetUintId(), spv::Op::OpCompositeExtract, uvec3_cast_id, 0);
|
||||
Instruction* v_inst = builder->AddIdLiteralOp(
|
||||
GetUintId(), spv::Op::OpCompositeExtract, uvec3_cast_id, 1);
|
||||
Instruction* u_inst =
|
||||
builder->AddCompositeExtract(GetUintId(), uvec3_cast_id, {0});
|
||||
Instruction* v_inst =
|
||||
builder->AddCompositeExtract(GetUintId(), uvec3_cast_id, {1});
|
||||
GenDebugOutputFieldCode(base_offset_id, kInstTessEvalOutTessCoordU,
|
||||
u_inst->result_id(), builder);
|
||||
GenDebugOutputFieldCode(base_offset_id, kInstTessEvalOutTessCoordV,
|
||||
@@ -302,8 +337,8 @@ void InstrumentPass::GenStageStreamWriteCode(uint32_t stage_idx,
|
||||
} break;
|
||||
case spv::ExecutionModel::Fragment: {
|
||||
// Load FragCoord and convert to Uint
|
||||
Instruction* frag_coord_inst = builder->AddUnaryOp(
|
||||
GetVec4FloatId(), spv::Op::OpLoad,
|
||||
Instruction* frag_coord_inst = builder->AddLoad(
|
||||
GetVec4FloatId(),
|
||||
context()->GetBuiltinInputVarId(uint32_t(spv::BuiltIn::FragCoord)));
|
||||
Instruction* uint_frag_coord_inst = builder->AddUnaryOp(
|
||||
GetVec4UintId(), spv::Op::OpBitcast, frag_coord_inst->result_id());
|
||||
@@ -321,12 +356,12 @@ void InstrumentPass::GenStageStreamWriteCode(uint32_t stage_idx,
|
||||
uint32_t launch_id = GenVarLoad(
|
||||
context()->GetBuiltinInputVarId(uint32_t(spv::BuiltIn::LaunchIdNV)),
|
||||
builder);
|
||||
Instruction* x_launch_inst = builder->AddIdLiteralOp(
|
||||
GetUintId(), spv::Op::OpCompositeExtract, launch_id, 0);
|
||||
Instruction* y_launch_inst = builder->AddIdLiteralOp(
|
||||
GetUintId(), spv::Op::OpCompositeExtract, launch_id, 1);
|
||||
Instruction* z_launch_inst = builder->AddIdLiteralOp(
|
||||
GetUintId(), spv::Op::OpCompositeExtract, launch_id, 2);
|
||||
Instruction* x_launch_inst =
|
||||
builder->AddCompositeExtract(GetUintId(), launch_id, {0});
|
||||
Instruction* y_launch_inst =
|
||||
builder->AddCompositeExtract(GetUintId(), launch_id, {1});
|
||||
Instruction* z_launch_inst =
|
||||
builder->AddCompositeExtract(GetUintId(), launch_id, {2});
|
||||
GenDebugOutputFieldCode(base_offset_id, kInstRayTracingOutLaunchIdX,
|
||||
x_launch_inst->result_id(), builder);
|
||||
GenDebugOutputFieldCode(base_offset_id, kInstRayTracingOutLaunchIdY,
|
||||
@@ -344,11 +379,10 @@ void InstrumentPass::GenDebugStreamWrite(
|
||||
// Call debug output function. Pass func_idx, instruction_idx and
|
||||
// validation ids as args.
|
||||
uint32_t val_id_cnt = static_cast<uint32_t>(validation_ids.size());
|
||||
uint32_t output_func_id = GetStreamWriteFunctionId(stage_idx, val_id_cnt);
|
||||
std::vector<uint32_t> args = {output_func_id,
|
||||
builder->GetUintConstantId(instruction_idx)};
|
||||
std::vector<uint32_t> args = {builder->GetUintConstantId(instruction_idx)};
|
||||
(void)args.insert(args.end(), validation_ids.begin(), validation_ids.end());
|
||||
(void)builder->AddNaryOp(GetVoidId(), spv::Op::OpFunctionCall, args);
|
||||
(void)builder->AddFunctionCall(
|
||||
GetVoidId(), GetStreamWriteFunctionId(stage_idx, val_id_cnt), args);
|
||||
}
|
||||
|
||||
bool InstrumentPass::AllConstant(const std::vector<uint32_t>& ids) {
|
||||
@@ -360,33 +394,38 @@ bool InstrumentPass::AllConstant(const std::vector<uint32_t>& ids) {
|
||||
}
|
||||
|
||||
uint32_t InstrumentPass::GenDebugDirectRead(
|
||||
const std::vector<uint32_t>& offset_ids, InstructionBuilder* ref_builder) {
|
||||
const std::vector<uint32_t>& offset_ids, InstructionBuilder* builder) {
|
||||
// Call debug input function. Pass func_idx and offset ids as args.
|
||||
uint32_t off_id_cnt = static_cast<uint32_t>(offset_ids.size());
|
||||
uint32_t input_func_id = GetDirectReadFunctionId(off_id_cnt);
|
||||
std::vector<uint32_t> args = {input_func_id};
|
||||
(void)args.insert(args.end(), offset_ids.begin(), offset_ids.end());
|
||||
const uint32_t off_id_cnt = static_cast<uint32_t>(offset_ids.size());
|
||||
const uint32_t input_func_id = GetDirectReadFunctionId(off_id_cnt);
|
||||
return GenReadFunctionCall(input_func_id, offset_ids, builder);
|
||||
}
|
||||
|
||||
uint32_t InstrumentPass::GenReadFunctionCall(
|
||||
uint32_t func_id, const std::vector<uint32_t>& func_call_args,
|
||||
InstructionBuilder* ref_builder) {
|
||||
// If optimizing direct reads and the call has already been generated,
|
||||
// use its result
|
||||
if (opt_direct_reads_) {
|
||||
uint32_t res_id = call2id_[args];
|
||||
uint32_t res_id = call2id_[func_call_args];
|
||||
if (res_id != 0) return res_id;
|
||||
}
|
||||
// If the offsets are all constants, the call can be moved to the first block
|
||||
// of the function where its result can be reused. One example where this is
|
||||
// profitable is for uniform buffer references, of which there are often many.
|
||||
// If the function arguments are all constants, the call can be moved to the
|
||||
// first block of the function where its result can be reused. One example
|
||||
// where this is profitable is for uniform buffer references, of which there
|
||||
// are often many.
|
||||
InstructionBuilder builder(ref_builder->GetContext(),
|
||||
&*ref_builder->GetInsertPoint(),
|
||||
ref_builder->GetPreservedAnalysis());
|
||||
bool insert_in_first_block = opt_direct_reads_ && AllConstant(offset_ids);
|
||||
bool insert_in_first_block = opt_direct_reads_ && AllConstant(func_call_args);
|
||||
if (insert_in_first_block) {
|
||||
Instruction* insert_before = &*curr_func_->begin()->tail();
|
||||
builder.SetInsertPoint(insert_before);
|
||||
}
|
||||
uint32_t res_id =
|
||||
builder.AddNaryOp(GetUintId(), spv::Op::OpFunctionCall, args)
|
||||
builder.AddFunctionCall(GetUintId(), func_id, func_call_args)
|
||||
->result_id();
|
||||
if (insert_in_first_block) call2id_[args] = res_id;
|
||||
if (insert_in_first_block) call2id_[func_call_args] = res_id;
|
||||
return res_id;
|
||||
}
|
||||
|
||||
@@ -502,32 +541,61 @@ uint32_t InstrumentPass::GetInputBufferBinding() {
|
||||
return 0;
|
||||
}
|
||||
|
||||
analysis::Type* InstrumentPass::GetUintXRuntimeArrayType(
|
||||
uint32_t width, analysis::Type** rarr_ty) {
|
||||
analysis::Integer* InstrumentPass::GetInteger(uint32_t width, bool is_signed) {
|
||||
analysis::Integer i(width, is_signed);
|
||||
analysis::Type* type = context()->get_type_mgr()->GetRegisteredType(&i);
|
||||
assert(type && type->AsInteger());
|
||||
return type->AsInteger();
|
||||
}
|
||||
|
||||
analysis::Struct* InstrumentPass::GetStruct(
|
||||
const std::vector<const analysis::Type*>& fields) {
|
||||
analysis::Struct s(fields);
|
||||
analysis::Type* type = context()->get_type_mgr()->GetRegisteredType(&s);
|
||||
assert(type && type->AsStruct());
|
||||
return type->AsStruct();
|
||||
}
|
||||
|
||||
analysis::RuntimeArray* InstrumentPass::GetRuntimeArray(
|
||||
const analysis::Type* element) {
|
||||
analysis::RuntimeArray r(element);
|
||||
analysis::Type* type = context()->get_type_mgr()->GetRegisteredType(&r);
|
||||
assert(type && type->AsRuntimeArray());
|
||||
return type->AsRuntimeArray();
|
||||
}
|
||||
|
||||
analysis::Function* InstrumentPass::GetFunction(
|
||||
const analysis::Type* return_val,
|
||||
const std::vector<const analysis::Type*>& args) {
|
||||
analysis::Function func(return_val, args);
|
||||
analysis::Type* type = context()->get_type_mgr()->GetRegisteredType(&func);
|
||||
assert(type && type->AsFunction());
|
||||
return type->AsFunction();
|
||||
}
|
||||
|
||||
analysis::RuntimeArray* InstrumentPass::GetUintXRuntimeArrayType(
|
||||
uint32_t width, analysis::RuntimeArray** rarr_ty) {
|
||||
if (*rarr_ty == nullptr) {
|
||||
analysis::DecorationManager* deco_mgr = get_decoration_mgr();
|
||||
analysis::TypeManager* type_mgr = context()->get_type_mgr();
|
||||
analysis::Integer uint_ty(width, false);
|
||||
analysis::Type* reg_uint_ty = type_mgr->GetRegisteredType(&uint_ty);
|
||||
analysis::RuntimeArray uint_rarr_ty_tmp(reg_uint_ty);
|
||||
*rarr_ty = type_mgr->GetRegisteredType(&uint_rarr_ty_tmp);
|
||||
uint32_t uint_arr_ty_id = type_mgr->GetTypeInstruction(*rarr_ty);
|
||||
*rarr_ty = GetRuntimeArray(GetInteger(width, false));
|
||||
uint32_t uint_arr_ty_id =
|
||||
context()->get_type_mgr()->GetTypeInstruction(*rarr_ty);
|
||||
// By the Vulkan spec, a pre-existing RuntimeArray of uint must be part of
|
||||
// a block, and will therefore be decorated with an ArrayStride. Therefore
|
||||
// the undecorated type returned here will not be pre-existing and can
|
||||
// safely be decorated. Since this type is now decorated, it is out of
|
||||
// sync with the TypeManager and therefore the TypeManager must be
|
||||
// invalidated after this pass.
|
||||
assert(context()->get_def_use_mgr()->NumUses(uint_arr_ty_id) == 0 &&
|
||||
assert(get_def_use_mgr()->NumUses(uint_arr_ty_id) == 0 &&
|
||||
"used RuntimeArray type returned");
|
||||
deco_mgr->AddDecorationVal(
|
||||
get_decoration_mgr()->AddDecorationVal(
|
||||
uint_arr_ty_id, uint32_t(spv::Decoration::ArrayStride), width / 8u);
|
||||
}
|
||||
return *rarr_ty;
|
||||
}
|
||||
|
||||
analysis::Type* InstrumentPass::GetUintRuntimeArrayType(uint32_t width) {
|
||||
analysis::Type** rarr_ty =
|
||||
analysis::RuntimeArray* InstrumentPass::GetUintRuntimeArrayType(
|
||||
uint32_t width) {
|
||||
analysis::RuntimeArray** rarr_ty =
|
||||
(width == 64) ? &uint64_rarr_ty_ : &uint32_rarr_ty_;
|
||||
return GetUintXRuntimeArrayType(width, rarr_ty);
|
||||
}
|
||||
@@ -546,11 +614,10 @@ uint32_t InstrumentPass::GetOutputBufferId() {
|
||||
// If not created yet, create one
|
||||
analysis::DecorationManager* deco_mgr = get_decoration_mgr();
|
||||
analysis::TypeManager* type_mgr = context()->get_type_mgr();
|
||||
analysis::Type* reg_uint_rarr_ty = GetUintRuntimeArrayType(32);
|
||||
analysis::Integer uint_ty(32, false);
|
||||
analysis::Type* reg_uint_ty = type_mgr->GetRegisteredType(&uint_ty);
|
||||
analysis::Struct buf_ty({reg_uint_ty, reg_uint_ty, reg_uint_rarr_ty});
|
||||
analysis::Type* reg_buf_ty = type_mgr->GetRegisteredType(&buf_ty);
|
||||
analysis::RuntimeArray* reg_uint_rarr_ty = GetUintRuntimeArrayType(32);
|
||||
analysis::Integer* reg_uint_ty = GetInteger(32, false);
|
||||
analysis::Type* reg_buf_ty =
|
||||
GetStruct({reg_uint_ty, reg_uint_ty, reg_uint_rarr_ty});
|
||||
uint32_t obufTyId = type_mgr->GetTypeInstruction(reg_buf_ty);
|
||||
// By the Vulkan spec, a pre-existing struct containing a RuntimeArray
|
||||
// must be a block, and will therefore be decorated with Block. Therefore
|
||||
@@ -604,8 +671,7 @@ uint32_t InstrumentPass::GetInputBufferId() {
|
||||
analysis::TypeManager* type_mgr = context()->get_type_mgr();
|
||||
uint32_t width = (validation_id_ == kInstValidationIdBuffAddr) ? 64u : 32u;
|
||||
analysis::Type* reg_uint_rarr_ty = GetUintRuntimeArrayType(width);
|
||||
analysis::Struct buf_ty({reg_uint_rarr_ty});
|
||||
analysis::Type* reg_buf_ty = type_mgr->GetRegisteredType(&buf_ty);
|
||||
analysis::Struct* reg_buf_ty = GetStruct({reg_uint_rarr_ty});
|
||||
uint32_t ibufTyId = type_mgr->GetTypeInstruction(reg_buf_ty);
|
||||
// By the Vulkan spec, a pre-existing struct containing a RuntimeArray
|
||||
// must be a block, and will therefore be decorated with Block. Therefore
|
||||
@@ -747,37 +813,17 @@ uint32_t InstrumentPass::GetStreamWriteFunctionId(uint32_t stage_idx,
|
||||
// Create function
|
||||
param2output_func_id_[param_cnt] = TakeNextId();
|
||||
analysis::TypeManager* type_mgr = context()->get_type_mgr();
|
||||
std::vector<const analysis::Type*> param_types;
|
||||
for (uint32_t c = 0; c < param_cnt; ++c)
|
||||
param_types.push_back(type_mgr->GetType(GetUintId()));
|
||||
analysis::Function func_ty(type_mgr->GetType(GetVoidId()), param_types);
|
||||
analysis::Type* reg_func_ty = type_mgr->GetRegisteredType(&func_ty);
|
||||
std::unique_ptr<Instruction> func_inst(
|
||||
new Instruction(get_module()->context(), spv::Op::OpFunction,
|
||||
GetVoidId(), param2output_func_id_[param_cnt],
|
||||
{{spv_operand_type_t::SPV_OPERAND_TYPE_LITERAL_INTEGER,
|
||||
{uint32_t(spv::FunctionControlMask::MaskNone)}},
|
||||
{spv_operand_type_t::SPV_OPERAND_TYPE_ID,
|
||||
{type_mgr->GetTypeInstruction(reg_func_ty)}}}));
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*func_inst);
|
||||
std::unique_ptr<Function> output_func =
|
||||
MakeUnique<Function>(std::move(func_inst));
|
||||
// Add parameters
|
||||
std::vector<uint32_t> param_vec;
|
||||
for (uint32_t c = 0; c < param_cnt; ++c) {
|
||||
uint32_t pid = TakeNextId();
|
||||
param_vec.push_back(pid);
|
||||
std::unique_ptr<Instruction> param_inst(
|
||||
new Instruction(get_module()->context(), spv::Op::OpFunctionParameter,
|
||||
GetUintId(), pid, {}));
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*param_inst);
|
||||
output_func->AddParameter(std::move(param_inst));
|
||||
}
|
||||
|
||||
const std::vector<const analysis::Type*> param_types(param_cnt,
|
||||
GetInteger(32, false));
|
||||
std::unique_ptr<Function> output_func = StartFunction(
|
||||
param2output_func_id_[param_cnt], type_mgr->GetVoidType(), param_types);
|
||||
|
||||
std::vector<uint32_t> param_ids = AddParameters(*output_func, param_types);
|
||||
|
||||
// Create first block
|
||||
uint32_t test_blk_id = TakeNextId();
|
||||
std::unique_ptr<Instruction> test_label(NewLabel(test_blk_id));
|
||||
std::unique_ptr<BasicBlock> new_blk_ptr =
|
||||
MakeUnique<BasicBlock>(std::move(test_label));
|
||||
auto new_blk_ptr = MakeUnique<BasicBlock>(NewLabel(TakeNextId()));
|
||||
|
||||
InstructionBuilder builder(
|
||||
context(), &*new_blk_ptr,
|
||||
IRContext::kAnalysisDefUse | IRContext::kAnalysisInstrToBlockMapping);
|
||||
@@ -786,9 +832,9 @@ uint32_t InstrumentPass::GetStreamWriteFunctionId(uint32_t stage_idx,
|
||||
uint32_t obuf_record_sz = val_spec_offset + val_spec_param_cnt;
|
||||
uint32_t buf_id = GetOutputBufferId();
|
||||
uint32_t buf_uint_ptr_id = GetOutputBufferPtrId();
|
||||
Instruction* obuf_curr_sz_ac_inst =
|
||||
builder.AddBinaryOp(buf_uint_ptr_id, spv::Op::OpAccessChain, buf_id,
|
||||
builder.GetUintConstantId(kDebugOutputSizeOffset));
|
||||
Instruction* obuf_curr_sz_ac_inst = builder.AddAccessChain(
|
||||
buf_uint_ptr_id, buf_id,
|
||||
{builder.GetUintConstantId(kDebugOutputSizeOffset)});
|
||||
// Fetch the current debug buffer written size atomically, adding the
|
||||
// size of the record to be written.
|
||||
uint32_t obuf_record_sz_id = builder.GetUintConstantId(obuf_record_sz);
|
||||
@@ -802,8 +848,8 @@ uint32_t InstrumentPass::GetStreamWriteFunctionId(uint32_t stage_idx,
|
||||
uint32_t obuf_curr_sz_id = obuf_curr_sz_inst->result_id();
|
||||
// Compute new written size
|
||||
Instruction* obuf_new_sz_inst =
|
||||
builder.AddBinaryOp(GetUintId(), spv::Op::OpIAdd, obuf_curr_sz_id,
|
||||
builder.GetUintConstantId(obuf_record_sz));
|
||||
builder.AddIAdd(GetUintId(), obuf_curr_sz_id,
|
||||
builder.GetUintConstantId(obuf_record_sz));
|
||||
// Fetch the data bound
|
||||
Instruction* obuf_bnd_inst =
|
||||
builder.AddIdLiteralOp(GetUintId(), spv::Op::OpArrayLength,
|
||||
@@ -825,13 +871,13 @@ uint32_t InstrumentPass::GetStreamWriteFunctionId(uint32_t stage_idx,
|
||||
new_blk_ptr = MakeUnique<BasicBlock>(std::move(write_label));
|
||||
builder.SetInsertPoint(&*new_blk_ptr);
|
||||
// Generate common and stage-specific debug record members
|
||||
GenCommonStreamWriteCode(obuf_record_sz, param_vec[kInstCommonParamInstIdx],
|
||||
GenCommonStreamWriteCode(obuf_record_sz, param_ids[kInstCommonParamInstIdx],
|
||||
stage_idx, obuf_curr_sz_id, &builder);
|
||||
GenStageStreamWriteCode(stage_idx, obuf_curr_sz_id, &builder);
|
||||
// Gen writes of validation specific data
|
||||
for (uint32_t i = 0; i < val_spec_param_cnt; ++i) {
|
||||
GenDebugOutputFieldCode(obuf_curr_sz_id, val_spec_offset + i,
|
||||
param_vec[kInstCommonParamCnt + i], &builder);
|
||||
param_ids[kInstCommonParamCnt + i], &builder);
|
||||
}
|
||||
// Close write block and gen merge block
|
||||
(void)builder.AddBranch(merge_blk_id);
|
||||
@@ -840,11 +886,9 @@ uint32_t InstrumentPass::GetStreamWriteFunctionId(uint32_t stage_idx,
|
||||
builder.SetInsertPoint(&*new_blk_ptr);
|
||||
// Close merge block and function and add function to module
|
||||
(void)builder.AddNullaryOp(0, spv::Op::OpReturn);
|
||||
|
||||
output_func->AddBasicBlock(std::move(new_blk_ptr));
|
||||
std::unique_ptr<Instruction> func_end_inst(new Instruction(
|
||||
get_module()->context(), spv::Op::OpFunctionEnd, 0, 0, {}));
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*func_end_inst);
|
||||
output_func->SetFunctionEnd(std::move(func_end_inst));
|
||||
output_func->SetFunctionEnd(EndFunction());
|
||||
context()->AddFunction(std::move(output_func));
|
||||
|
||||
std::string name("stream_write_");
|
||||
@@ -861,77 +905,48 @@ uint32_t InstrumentPass::GetDirectReadFunctionId(uint32_t param_cnt) {
|
||||
if (func_id != 0) return func_id;
|
||||
// Create input function for param_cnt.
|
||||
func_id = TakeNextId();
|
||||
analysis::TypeManager* type_mgr = context()->get_type_mgr();
|
||||
std::vector<const analysis::Type*> param_types;
|
||||
for (uint32_t c = 0; c < param_cnt; ++c)
|
||||
param_types.push_back(type_mgr->GetType(GetUintId()));
|
||||
uint32_t ibuf_type_id = GetInputBufferTypeId();
|
||||
analysis::Function func_ty(type_mgr->GetType(ibuf_type_id), param_types);
|
||||
analysis::Type* reg_func_ty = type_mgr->GetRegisteredType(&func_ty);
|
||||
std::unique_ptr<Instruction> func_inst(new Instruction(
|
||||
get_module()->context(), spv::Op::OpFunction, ibuf_type_id, func_id,
|
||||
{{spv_operand_type_t::SPV_OPERAND_TYPE_LITERAL_INTEGER,
|
||||
{uint32_t(spv::FunctionControlMask::MaskNone)}},
|
||||
{spv_operand_type_t::SPV_OPERAND_TYPE_ID,
|
||||
{type_mgr->GetTypeInstruction(reg_func_ty)}}}));
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*func_inst);
|
||||
analysis::Integer* uint_type = GetInteger(32, false);
|
||||
std::vector<const analysis::Type*> param_types(param_cnt, uint_type);
|
||||
|
||||
std::unique_ptr<Function> input_func =
|
||||
MakeUnique<Function>(std::move(func_inst));
|
||||
// Add parameters
|
||||
std::vector<uint32_t> param_vec;
|
||||
for (uint32_t c = 0; c < param_cnt; ++c) {
|
||||
uint32_t pid = TakeNextId();
|
||||
param_vec.push_back(pid);
|
||||
std::unique_ptr<Instruction> param_inst(
|
||||
new Instruction(get_module()->context(), spv::Op::OpFunctionParameter,
|
||||
GetUintId(), pid, {}));
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*param_inst);
|
||||
input_func->AddParameter(std::move(param_inst));
|
||||
}
|
||||
StartFunction(func_id, uint_type, param_types);
|
||||
std::vector<uint32_t> param_ids = AddParameters(*input_func, param_types);
|
||||
|
||||
// Create block
|
||||
uint32_t blk_id = TakeNextId();
|
||||
std::unique_ptr<Instruction> blk_label(NewLabel(blk_id));
|
||||
std::unique_ptr<BasicBlock> new_blk_ptr =
|
||||
MakeUnique<BasicBlock>(std::move(blk_label));
|
||||
auto new_blk_ptr = MakeUnique<BasicBlock>(NewLabel(TakeNextId()));
|
||||
InstructionBuilder builder(
|
||||
context(), &*new_blk_ptr,
|
||||
IRContext::kAnalysisDefUse | IRContext::kAnalysisInstrToBlockMapping);
|
||||
// For each offset parameter, generate new offset with parameter, adding last
|
||||
// loaded value if it exists, and load value from input buffer at new offset.
|
||||
// Return last loaded value.
|
||||
uint32_t ibuf_type_id = GetInputBufferTypeId();
|
||||
uint32_t buf_id = GetInputBufferId();
|
||||
uint32_t buf_ptr_id = GetInputBufferPtrId();
|
||||
uint32_t last_value_id = 0;
|
||||
for (uint32_t p = 0; p < param_cnt; ++p) {
|
||||
uint32_t offset_id;
|
||||
if (p == 0) {
|
||||
offset_id = param_vec[0];
|
||||
offset_id = param_ids[0];
|
||||
} else {
|
||||
if (ibuf_type_id != GetUintId()) {
|
||||
Instruction* ucvt_inst =
|
||||
builder.AddUnaryOp(GetUintId(), spv::Op::OpUConvert, last_value_id);
|
||||
last_value_id = ucvt_inst->result_id();
|
||||
last_value_id =
|
||||
builder.AddUnaryOp(GetUintId(), spv::Op::OpUConvert, last_value_id)
|
||||
->result_id();
|
||||
}
|
||||
Instruction* offset_inst = builder.AddBinaryOp(
|
||||
GetUintId(), spv::Op::OpIAdd, last_value_id, param_vec[p]);
|
||||
offset_id = offset_inst->result_id();
|
||||
offset_id = builder.AddIAdd(GetUintId(), last_value_id, param_ids[p])
|
||||
->result_id();
|
||||
}
|
||||
Instruction* ac_inst = builder.AddTernaryOp(
|
||||
buf_ptr_id, spv::Op::OpAccessChain, buf_id,
|
||||
builder.GetUintConstantId(kDebugInputDataOffset), offset_id);
|
||||
Instruction* load_inst =
|
||||
builder.AddUnaryOp(ibuf_type_id, spv::Op::OpLoad, ac_inst->result_id());
|
||||
last_value_id = load_inst->result_id();
|
||||
Instruction* ac_inst = builder.AddAccessChain(
|
||||
buf_ptr_id, buf_id,
|
||||
{builder.GetUintConstantId(kDebugInputDataOffset), offset_id});
|
||||
last_value_id =
|
||||
builder.AddLoad(ibuf_type_id, ac_inst->result_id())->result_id();
|
||||
}
|
||||
(void)builder.AddInstruction(MakeUnique<Instruction>(
|
||||
context(), spv::Op::OpReturnValue, 0, 0,
|
||||
std::initializer_list<Operand>{{SPV_OPERAND_TYPE_ID, {last_value_id}}}));
|
||||
(void)builder.AddUnaryOp(0, spv::Op::OpReturnValue, last_value_id);
|
||||
// Close block and function and add function to module
|
||||
input_func->AddBasicBlock(std::move(new_blk_ptr));
|
||||
std::unique_ptr<Instruction> func_end_inst(new Instruction(
|
||||
get_module()->context(), spv::Op::OpFunctionEnd, 0, 0, {}));
|
||||
get_def_use_mgr()->AnalyzeInstDefUse(&*func_end_inst);
|
||||
input_func->SetFunctionEnd(std::move(func_end_inst));
|
||||
input_func->SetFunctionEnd(EndFunction());
|
||||
context()->AddFunction(std::move(input_func));
|
||||
|
||||
std::string name("direct_read_");
|
||||
|
||||
@@ -214,6 +214,10 @@ class InstrumentPass : public Pass {
|
||||
uint32_t GenDebugDirectRead(const std::vector<uint32_t>& offset_ids,
|
||||
InstructionBuilder* builder);
|
||||
|
||||
uint32_t GenReadFunctionCall(uint32_t func_id,
|
||||
const std::vector<uint32_t>& args,
|
||||
InstructionBuilder* builder);
|
||||
|
||||
// Generate code to convert integer |value_id| to 32bit, if needed. Return
|
||||
// an id to the 32bit equivalent.
|
||||
uint32_t Gen32BitCvtCode(uint32_t value_id, InstructionBuilder* builder);
|
||||
@@ -222,6 +226,15 @@ class InstrumentPass : public Pass {
|
||||
// Return an id to the Uint equivalent.
|
||||
uint32_t GenUintCastCode(uint32_t value_id, InstructionBuilder* builder);
|
||||
|
||||
std::unique_ptr<Function> StartFunction(
|
||||
uint32_t func_id, const analysis::Type* return_type,
|
||||
const std::vector<const analysis::Type*>& param_types);
|
||||
|
||||
std::vector<uint32_t> AddParameters(
|
||||
Function& func, const std::vector<const analysis::Type*>& param_types);
|
||||
|
||||
std::unique_ptr<Instruction> EndFunction();
|
||||
|
||||
// Return new label.
|
||||
std::unique_ptr<Instruction> NewLabel(uint32_t label_id);
|
||||
|
||||
@@ -253,12 +266,20 @@ class InstrumentPass : public Pass {
|
||||
// Return id for void type
|
||||
uint32_t GetVoidId();
|
||||
|
||||
// Return pointer to type for runtime array of uint
|
||||
analysis::Type* GetUintXRuntimeArrayType(uint32_t width,
|
||||
analysis::Type** rarr_ty);
|
||||
// Get registered type structures
|
||||
analysis::Integer* GetInteger(uint32_t width, bool is_signed);
|
||||
analysis::Struct* GetStruct(const std::vector<const analysis::Type*>& fields);
|
||||
analysis::RuntimeArray* GetRuntimeArray(const analysis::Type* element);
|
||||
analysis::Function* GetFunction(
|
||||
const analysis::Type* return_val,
|
||||
const std::vector<const analysis::Type*>& args);
|
||||
|
||||
// Return pointer to type for runtime array of uint
|
||||
analysis::Type* GetUintRuntimeArrayType(uint32_t width);
|
||||
analysis::RuntimeArray* GetUintXRuntimeArrayType(
|
||||
uint32_t width, analysis::RuntimeArray** rarr_ty);
|
||||
|
||||
// Return pointer to type for runtime array of uint
|
||||
analysis::RuntimeArray* GetUintRuntimeArrayType(uint32_t width);
|
||||
|
||||
// Return id for buffer uint type
|
||||
uint32_t GetOutputBufferPtrId();
|
||||
@@ -448,10 +469,10 @@ class InstrumentPass : public Pass {
|
||||
bool storage_buffer_ext_defined_;
|
||||
|
||||
// runtime array of uint type
|
||||
analysis::Type* uint64_rarr_ty_;
|
||||
analysis::RuntimeArray* uint64_rarr_ty_;
|
||||
|
||||
// runtime array of uint type
|
||||
analysis::Type* uint32_rarr_ty_;
|
||||
analysis::RuntimeArray* uint32_rarr_ty_;
|
||||
|
||||
// Pre-instrumentation same-block insts
|
||||
std::unordered_map<uint32_t, Instruction*> same_block_pre_;
|
||||
|
||||
92
3rdparty/spirv-tools/source/opt/optimizer.cpp
vendored
92
3rdparty/spirv-tools/source/opt/optimizer.cpp
vendored
@@ -1065,3 +1065,95 @@ Optimizer::PassToken CreateFixFuncCallArgumentsPass() {
|
||||
MakeUnique<opt::FixFuncCallArgumentsPass>());
|
||||
}
|
||||
} // namespace spvtools
|
||||
|
||||
extern "C" {
|
||||
|
||||
SPIRV_TOOLS_EXPORT spv_optimizer_t* spvOptimizerCreate(spv_target_env env) {
|
||||
return reinterpret_cast<spv_optimizer_t*>(new spvtools::Optimizer(env));
|
||||
}
|
||||
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerDestroy(spv_optimizer_t* optimizer) {
|
||||
delete reinterpret_cast<spvtools::Optimizer*>(optimizer);
|
||||
}
|
||||
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerSetMessageConsumer(
|
||||
spv_optimizer_t* optimizer, spv_message_consumer consumer) {
|
||||
reinterpret_cast<spvtools::Optimizer*>(optimizer)->
|
||||
SetMessageConsumer(
|
||||
[consumer](spv_message_level_t level, const char* source,
|
||||
const spv_position_t& position, const char* message) {
|
||||
return consumer(level, source, &position, message);
|
||||
});
|
||||
}
|
||||
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerRegisterLegalizationPasses(
|
||||
spv_optimizer_t* optimizer) {
|
||||
reinterpret_cast<spvtools::Optimizer*>(optimizer)->
|
||||
RegisterLegalizationPasses();
|
||||
}
|
||||
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerRegisterPerformancePasses(
|
||||
spv_optimizer_t* optimizer) {
|
||||
reinterpret_cast<spvtools::Optimizer*>(optimizer)->
|
||||
RegisterPerformancePasses();
|
||||
}
|
||||
|
||||
SPIRV_TOOLS_EXPORT void spvOptimizerRegisterSizePasses(
|
||||
spv_optimizer_t* optimizer) {
|
||||
reinterpret_cast<spvtools::Optimizer*>(optimizer)->RegisterSizePasses();
|
||||
}
|
||||
|
||||
SPIRV_TOOLS_EXPORT bool spvOptimizerRegisterPassFromFlag(
|
||||
spv_optimizer_t* optimizer, const char* flag)
|
||||
{
|
||||
return reinterpret_cast<spvtools::Optimizer*>(optimizer)->
|
||||
RegisterPassFromFlag(flag);
|
||||
}
|
||||
|
||||
SPIRV_TOOLS_EXPORT bool spvOptimizerRegisterPassesFromFlags(
|
||||
spv_optimizer_t* optimizer, const char** flags, const size_t flag_count) {
|
||||
std::vector<std::string> opt_flags;
|
||||
for (uint32_t i = 0; i < flag_count; i++) {
|
||||
opt_flags.emplace_back(flags[i]);
|
||||
}
|
||||
|
||||
return reinterpret_cast<spvtools::Optimizer*>(optimizer)->
|
||||
RegisterPassesFromFlags(opt_flags);
|
||||
}
|
||||
|
||||
SPIRV_TOOLS_EXPORT
|
||||
spv_result_t spvOptimizerRun(spv_optimizer_t* optimizer,
|
||||
const uint32_t* binary,
|
||||
const size_t word_count,
|
||||
spv_binary* optimized_binary,
|
||||
const spv_optimizer_options options) {
|
||||
std::vector<uint32_t> optimized;
|
||||
|
||||
if (!reinterpret_cast<spvtools::Optimizer*>(optimizer)->
|
||||
Run(binary, word_count, &optimized, options)) {
|
||||
return SPV_ERROR_INTERNAL;
|
||||
}
|
||||
|
||||
auto result_binary = new spv_binary_t();
|
||||
if (!result_binary) {
|
||||
*optimized_binary = nullptr;
|
||||
return SPV_ERROR_OUT_OF_MEMORY;
|
||||
}
|
||||
|
||||
result_binary->code = new uint32_t[optimized.size()];
|
||||
if (!result_binary->code) {
|
||||
delete result_binary;
|
||||
*optimized_binary = nullptr;
|
||||
return SPV_ERROR_OUT_OF_MEMORY;
|
||||
}
|
||||
result_binary->wordCount = optimized.size();
|
||||
|
||||
memcpy(result_binary->code, optimized.data(),
|
||||
optimized.size() * sizeof(uint32_t));
|
||||
|
||||
*optimized_binary = result_binary;
|
||||
|
||||
return SPV_SUCCESS;
|
||||
}
|
||||
|
||||
} // extern "C"
|
||||
|
||||
@@ -206,13 +206,14 @@ spv_result_t BitwisePass(ValidationState_t& _, const Instruction* inst) {
|
||||
<< spvOpcodeString(opcode);
|
||||
|
||||
const uint32_t base_type = _.GetOperandTypeId(inst, 2);
|
||||
const uint32_t base_dimension = _.GetDimension(base_type);
|
||||
const uint32_t result_dimension = _.GetDimension(result_type);
|
||||
|
||||
if (spv_result_t error = ValidateBaseType(_, inst, base_type)) {
|
||||
return error;
|
||||
}
|
||||
|
||||
const uint32_t base_dimension = _.GetDimension(base_type);
|
||||
const uint32_t result_dimension = _.GetDimension(result_type);
|
||||
|
||||
if (base_dimension != result_dimension)
|
||||
return _.diag(SPV_ERROR_INVALID_DATA, inst)
|
||||
<< "Expected Base dimension to be equal to Result Type "
|
||||
|
||||
Reference in New Issue
Block a user