MDL SDK 2020.0.1 (327300.3640): 30 Mar 2020

ABI compatible with the MDL SDK 2020.0.1 (327300.3640) binary release
(see https://developer.nvidia.com/mdl-sdk)

Added and Changed Features

• General

  • The standalone tools mdlm and i18n have been extended to support Unicode package/module
    names.
  • On Windows, we recommend Visual Studio 2017 or 2019. Visual Studio 2015 is still the minimum
    requirement.

• MDL SDK examples

  • example_dxr:
    • Enhanced dependency tracking when reloading materials to also update indirectly affected
      materials.
    • Only the compiled material hash is considered to detect reusable generated code which
      allows to reuse existing materials for structurally equal instances.
    • Skip format conversion for multi-scatter lookup data.
    • Added support for larger glTF scenes up to 2GB.
    • Improved cleanup when loading erroneous scenes.

Fixed Bugs

• General

  • Fixed handling of resources inside function bodies. Previously, these resources were not
    found under some conditions, causing black textures for instance.
  • Fixed too strict error checks for creation of function calls of the array index operator, the
    ternary operator, and the cast operator.
  • Fixed creation of variants without specifying any annotations where the annotations of the
    prototype were erroneously copied to the variants.
  • Fixed loading of string-based modules with absolute file paths for resources.
  • Fixed documentation of generated code interfaces: The results of eval functions are
    output-only, not in/out.

• MDL Compiler and Backends

  • Fixed a subtle bug in one of the code caches, which caused ignored argument changes under
    some complex conditions. Typically, boolean parameters were vulnerable, but could happen to
    parameters of any type.
  • Fixed MDL archive tool failures with Unicode package names. The MDL version of such archives
    is now automatically set to MDL 1.6 as lowest necessary version.
  • A bug in the resource handling was fixed that previously caused resources to be resolved and
    loaded more that once, possibly leading to failures if search paths had been changed in
    between.
  • Fixed the MDL core compiler's analysis pass. Some analysis info was computed but not
    annotated, causing JIT failures on functions that consists of a single expression body only.

MDL SDK 2020 (327300.2022): 22 Feb 2020

ABI compatible with the MDL SDK 2020 (327300.2022) binary release
(see https://developer.nvidia.com/mdl-sdk)

Added and Changed Features

• General

  • A new function mi::neuraylib::IMdl_compiler::get_df_data_texture() has been added.
  • A new function mi::neuraylib::ITarget_code::get_texture_df_data_kind() has been
    added.
  • A new enum mi::neuraylib::Df_data_kind has been added.
  • A new flag on mi::neuraylib::IMdl_configuration instructs the MDL compiler to keep the
    names of let expressions and expose them as temporaries on MDL material and function definitions.
    This brings the structure of the material/function definition presented in the API closer to the
    one in the .mdl file.
  • The FreeImage plugin is now based on FreeImage 3.18.0.

• MDL Compiler and Backends

  • Support for the chiang hair_bsdf has been added to the code generator for distribution
    functions.

  • Changes to the internal representation of builtin MDL operators.
    MDL supports a variety of operators, potentially featuring an endless number of instances:

    • array index operator[]
    • array length symbol
    • ternary operator ?:

    Previously, 'local' definitions were created for every used instance of these operators in an MDL
    module:

    • array index operator on type T in module M: M::T@(T,int)
    • array length symbol on type T in module M: M::T.len(T)
    • ternary operator on type T in module M: M::operator?(bool,T,T)

    This representation had several drawbacks:

    • there might be one definition for the same operator in every module
    • if the operator was not used inside the source of a module, it was not created

    Especially the second point lead to several problems in the editing application. Hence, starting
    with the 2020.0.0 release, the internal representation was changed and operators are now
    represented by 'global' template-like definitions:

    • array index operator: operator[](<0>[],int)
    • array length operator: operator_len(<0>[])
    • ternary operator: operator?(bool,<0>,<0>)

    In addition, the name of the cast operator was changed from operator_cast() to
    operator_cast(<0>).

    Drawback: When inspecting the types of the operators definition, 'int' is returned for the
    template types, but this might be changed in the future by expanding the type system.

  • Support for HLSL scene data renderer runtime functions has been added. See the
    scene_data_* functions in mdl_renderer_runtime.hlsl MDL SDK DXR example for an
    example implementation.

• MDL SDK examples

  • example_dxr:
    • The texture loading pipeline has been simplified and support for UDIM textures has been
      added.
    • Support for scene data introduced in MDL 1.6 (prim vars) has been added.
  • example_df_cuda:
    • Support for evaluation of hair-bsdfs on an analytical cylinder has been added.

Fixed Bugs

• MDL Compiler and Backends

  • Support for multiple multiscatter textures in one target code object has been fixed.

  • Support for multiscatter textures with disabled resolve_resources backend option has
    been fixed.

  • Multiple HLSL code generation problems leading to non-compilable code have been fixed.

MDL SDK 2019.2 (325000.1814): 18 Dec 2019

ABI compatible with the MDL SDK 2019.2 (325000.1814) binary release
(see https://developer.nvidia.com/mdl-sdk)

Added and Changed Features

• MDL 1.6 Language Specification

  • The file path resolution algorithm has been changed to treat weak relative
    paths the same as strict relative paths if the referring MDL
    module has an MDL version of 1.6 or higher. Furthermore,
    the error checks have been simplified to only protect relative paths from
    referring to files in other search paths.
  • The import of standard library modules has been changed in all examples
    to use absolute path imports.
  • An additional way of defining functions has been added using an expression
    instead of a procedural function body.
  • Let-expression can also be applied to functions defined using an expression.
  • The limitation has been removed that package names and module names can
    only be identifiers.
  • The new using alias declaration has been added to enable the use of Unicode
    names for module names and package names.
  • The description has been clarified that standard module names shadow only
    modules of the same fully qualified name while modules in subpackages can have
    a standard module name as their unqualified name.
  • The new scene standard library module has been added with
    data_isvalid, data_lookup_ltype, and
    data_lookup_uniform_ltype functions.
  • The new multiscatter_tint parameter has been added to all glossy BSDF
    models to enable energy loss compensation at higher roughness values.
  • The new df::sheen_bsdf bidirectional scattering distribution function has
    been added.
  • The new df::tint modifier overload has been added for the hair bidirectional
    scattering distribution function.
  • The new df::tint modifier overload has been added for the separate tinting
    of the reflective and transmissive light paths of a base BSDF.

• General

  • The new API functions
    • mi::neuraylib::IModule::reload()
    • mi::neuraylib::IModule::reload_from_string()
    • mi::neuraylib::IModule::is_valid()
    • mi::neuraylib::IMaterial_definition::is_valid()
    • mi::neuraylib::IFunction_definition::is_valid()
    • mi::neuraylib::IMaterial_instance::is_valid()
    • mi::neuraylib::IMaterial_instance::repair()
    • mi::neuraylib::IFunction_call::is_valid()
    • mi::neuraylib::IFunction_call::repair()
    • mi::neuraylib::ICompiled_material::is_valid()
      have been added to support reloading of MDL modules.
  • The requirements on MDL module names have been relaxed according to the MDL 1.6
    Specification to allow loading of modules with Unicode names.
  • The new API functions
    • mi::neuraylib::ITarget_code::get_callable_function_df_handle_count() and
    • mi::neuraylib::ITarget_code::get_callable_function_df_handle()
      have been added.
  • The new API function mi::neuraylib::ITarget_code::get_texture_df_data()
    has been added.
  • mi::neuraylib::IMdl_compiler::load_module() can now be called from multiple
    threads to allow loading modules in parallel. To support custom thread blocking
    the new interfaces
    • mi::neuraylib::IMdl_loading_wait_handle and
    • mi::neuraylib::IMdl_loading_wait_handle_factory
      have been added.
  • The new API functions
    • mi::neuraylib::IMaterial_definition::get_body()
    • mi::neuraylib::IMaterial_definition::get_temporary_count()
    • mi::neuraylib::IMaterial_definition::get_temporary()
    • mi::neuraylib::IFunction_definition::get_body()
    • mi::neuraylib::IFunction_definition::get_temporary_count()
    • mi::neuraylib::IFunction_definition::get_temporary()
      have been added.
  • The signature of the function mi::base::ILogger::message() has been changed.
  • The API function mi::neuraylib::ITransaction::edit() has been adapted to disallow
    editing of database elements of type mi::neuraylib::IMaterial_definition and
    mi::neuraylib::IFunction_definition.
  • Support for multiple occurrence of the same annotation has been added to
    mi::neuraylib::Annotation_wrapper.
  • Support for deprecated features guarded by MI_NEURAYLIB_DEPRECATED_8_1 and
    MI_NEURAYLIB_DEPRECATED_9_1 has been removed.

• MDL Compiler and Backends

  • Support for MDL 1.6 has been added to the MDL core compiler.
  • Limited support for MDL 1.6 features has been added to the backends, in
    particular, the scene module is supported, but currently no code is generated
    for interrogating the renderer, hence always the default value is returned.
  • Support for MDL 1.6 has been added to the generated code for distribution
    functions, that is df::tint(reflection_tint, transmission_tint),
    df::sheen_bsdf(), the multiscatter_tint parameter of all BSDFs exposing this
    in MDL 1.6 (note that this requires that all four dimensions of
    Bsdf_sample_data::xi are set).
  • For evaluating parts of distribution functions that are named by handles,
    Bsdf_evaluate_data and Bsdf_auxiliary_data are adapted to select individual
    handles.
  • A new backend option df_handle_slot_mode to select how evaluate and auxiliary
    data is passed between the generated code and the render has been added.
  • The bsdf field of Bsdf_evaluate_data is split into bsdf_diffuse and
    bsdf_glossy.
  • The Bsdf_sample_data structure now requires a 4th uniform random number and
    returns the handle of the sampled distribution part.
  • Inlining of functions containing constant declarations into the DAG has been
    implemented.
  • Support for light-path-expressions in generated code for distribution functions
    via handles has been added.
  • Support for retrieving albedo and normal in generated code for distribution
    functions via generated auxiliary functions has been added.
  • The entity resolver has been sped up for built-in modules in some cases where it
    is clear that the module can only be read from the MDL root.
  • The memory size of the DAG representation has been slightly reduced by
    internalizing all DAG signatures.
  • The DAG representation now uses unsafe math operations, especially x * 0 = 0
    for floating point values.

• MDL SDK examples

  • Example df_cuda has been adapted to illustrate how to evaluate parts of the
    distribution functions named by handles via light path expressions (LPEs).
  • All examples have been adapted to support processing of Unicode command line
    arguments.

Fixed Bugs

• General

  • An issue in the light profile parser has been fixed: For IESNA LM-63-2002 files
    the ballast-lamp value incorrectly acted as multiplier for the intensity.

• MDL Compiler and Backends

  • Several issues in the generated code for distribution functions have been fixed:
    • Bugs in the computation of the pdf and eval functions of
      df::ward_geisler_moroder_bsdf have been fixed.
    • Incorrect pdf computation (for sample, eval, and pdf functions) in
      df::ward_geisler_moroder_bsdf and df::backscattering_glossy_reflection_bsdf
      have been fixed.
    • Fixed a missing re-scale of pseudorandom numbers for v-cavities based masking,
      leading to biased results for df::scatter_reflect_transmit.
    • Only use refraction-based half vector for Fresnel-layering, not for all curve
      layering operations.
    • Add simple inside/outside material support based on IOR comparison to determine
      which IOR to override in Fresnel layering. This fixes incorrect rendering when
      BSDFs of type df::scatter_reflect and df::scatter_transmit are layered using
      df::fresnel_layer, in particular missing total internal reflection.
  • The implementation of math::isnan() and math::isfinite() has been fixed for
    vector types.
  • Printing of quiet NaNs for HLSL has been fixed.
  • A crash in the MDL core compiler that could occur if exported types contain errors
    in their default initializers has been fixed.
  • Wrong function names generated from debug::assert() calls when placed after a
    while loop have been fixed.
  • The name of the anno::deprecated() parameter has been fixed, it is description,
    not message.
  • The export of MDL modules containing relative imports has been fixed, access to
    the imported entities is now generated correctly.

MDL SDK 2019.1.4 (317500.5028): 27 Aug 2019

ABI compatible with the MDL SDK 2019.1.4 (317500.5028) binary release
(see https://developer.nvidia.com/mdl-sdk)

Added and Changed Features

• General

  • A new function mi::neuraylib::IValue_texture::get_gamma() has been added.
  • The following new functions have been added to the target code generation:
    • mi::neuraylib::ITarget_code::execute_bsdf_auxiliary()
    • mi::neuraylib::ITarget_code::execute_edf_auxiliary()
  • A new function mi::neuraylib::ICompiled_material::get_surface_opacity()
    has been added.

• MDL Compiler and Backends

  • Code generation for auxiliary methods has been added on distribution
    functions for potential use in AI-denoising.
  • The spectral color constructor color(float[<N>],float[N]),
    math::emission_color(), and math::blackboby() are now supported in the
    JIT backend.
  • More optimizations regarding elemental constructors in the DAG
    representation have been implemented.
  • Map XOR operators on Boolean values to NOT-EQUAL in the HLSL backend to be
    compatible to the SLANG compiler.

• MDL SDK examples

  • The example programs example_dxr and example_df_cuda have been extended
    to illustrate the use of auxiliary functions.
  • A modified version of example_dxr has been added to illustrate the usage
    of MDL in a multi-threaded context.
  • Camera controls have been improved and new options have been added to
    the example program example_dxr.

Fixed Bugs

• General

  • The export of MDLE files from in-memory MDL modules has been fixed.

• MDL Compiler and Backends

  • Temporary exponential creation of DAG nodes when using derivatives has
    been fixed.
  • Code generation of parameters reused multiple times in a derivative context
    has been fixed.
  • Relative imports including "." and ".." have been fixed.
  • Duplicate global variables in generated HLSL code have been fixed.
  • Invalid code generation for HLSL for special materials has been fixed.
  • Indeterministic rare compilation errors regarding unknown functions have
    been fixed.
  • Indeterministic rare hangs during compilation with multiple threads have
    been fixed.
  • Under rare condition the code cache could return HLSL code instead of PTX
    and vice versa. This has been fixed.
  • The code cache that was not working under several conditions has been fixed.
  • The handling of the ?: operator on arrays inside the DAG representation
    has been fixed such that it computes the right name now.
  • The handling of unresolved resource paths in the target code has been fixed.
    Previously all resources were mapped to index 1.
  • A crash in the code generator when handling uniform matrix expressions with
    automatic derivatives enabled has been fixed.
  • A crash in the HLSL code generator for non-default optimization levels has
    been fixed.
  • A crash when adding distribution functions to a link unit after adding non-
    distribution functions has been fixed.

MDL SDK 2019.1.1 (317500.2554): 08 Jun 2019

ABI compatible with the MDL SDK 2019.1.1 (317500.2554) binary release
(see https://developer.nvidia.com/mdl-sdk)

Added and Changed Features

• General

  • A new API class mi::neuraylib::IAnnotation_definition has been added.
  • The following new functions have been added:
    • mi::neuraylib::IAnnotation::get_definition()
    • mi::neuraylib::IModule::get_annotation_definition_count()
    • mi::neuraylib::IModule::get_annotation_definition(Size)
    • mi::neuraylib::IModule::get_annotation_definition(const char*)
  • Added the Boolean option fold_ternary_on_df to fold ternary operators on
    distribution function types in material bodies even in class compilation mode:
    Some uber-materials extensively use the ternary operator to switch BSDFs. This
    causes a lot of generated code and may not be supported by some renderers.
    Enabling this option will compile all arguments that control the condition of
    such a ternary operator in. Changing any of these arguments will then require
    a recompilation.

• MDL SDK examples

  • Support for the glTF extension KHR_materials_pbrSpecularGlossiness has been
    added to example_dxr.

Fixed Bugs

• General

  • Wrong copying of gamma values during cloning of texture values has been fixed.
    This especially happened for default parameters during material instantiation.
  • Fixed problem with dangling user defined types when a module is removed;
    previously removing a module left all its user defined types alive, preventing
    creating new (different) ones when the deleted module was reloaded.
  • Fixed failure when generating MDLE files that reference intensity_mode values.

• MDL Compiler and Backends

  • Fixed wrong optimization of empty do-while loops in the core compiler.
  • Fixed imports starting with . or .., which caused wrong package names before.
  • Fixed printing of float and double constants when MDL code was generated; not
    enough digits were used previously, causing lost precision when this code was
    compiled again.
  • Fixed a problem where sometimes several uninitialized variables were generated and
    copied around in generated HLSL code.
  • Fixed generation of useless copies, like t = t; in generated HLSL code.
  • Generated better HLSL code for vector constructor calls, like
    floatX(a.x, x.y, ...) => a.xy...
  • JIT compilation of functions involving sin and cos with the same argument on
    Mac OS has been fixed.
  • The implementation of df::color_weighted_layer has been fixed.

• MDL SDK examples

  • findGLEW has been fixed in the build script to work with CMake 3.15.

MDL SDK 2019.1 (317500.1752): 16 May 2019

ABI compatible with the MDL SDK 2019.1 (317500.1752) binary release
(see https://developer.nvidia.com/mdl-sdk)

Added and Changed Features

• MDL 1.5 Language Specification

  • A new cast operator has been added to support assignments between
    structurally equivalent user defined structure types and value
    equivalent enumeration types to support workflows with the new
    MDLE format. Beginning with MDL 1.5, cast is a reserved word.
  • A new field named hair of type hair_bsdf has been added to the
    material type, which represents the shading model applicable for hair
    primitives. Beginning with MDL 1.5, hair_bsdf is a reserved word.
  • A new elemental distribution function df::chiang_hair_bsdf has been
    added as a hair shading model.
  • A new distribution function modifier df::measured_factor has been
    added to support microfacet coloring based on the angle between the
    half-vector and the shading normal in addition to the angle between
    the half-vector and the incoming ray direction.
  • Annotations have been added to annotation declarations.
  • A new standard annotation origin() has been added, which is used in
    the MDLE file format to reference the original declarations of
    refactored elements.
  • The new Appendix D -- MDLE File Format defines a new container format
    for a self contained MDL material or function including all of its
    dependencies and resources.
  • The new Appendix E -- Internationalization defines the use of XLIFF
    files for the localization of MDL string annotations.

• General

  • A new function IMdle_api::get_hash() has been added.
  • A new function IMdl_compiler::get_module_db_name() has been added.
  • The MDLE file format version has been bumped to 1.0.
  • MDLE files now use the new anno::origin annotation rather than a custom one.
  • A new interface mi::neuraylib::IValue_string_localized has been added.
  • A new function IType_factory::is_compatible() has been added to check if one MDL
    type can be cast to another.
  • A new function IExpression_factory::create_cast() has been added.
  • A new configuration interface IMdl_configuration has been added, which can be used
    to control the behavior of the SDK regarding the automatic insertion of casts when
    assigning compatible but different types to arguments of MDL instances.
  • The IMdl_discovery_api has been extended to also support discovery of resources and
    XLIFF files.

• MDL Compiler and Backends

  • A new backend mi::neuraylib::IMdl_compiler::MB_HLSL for HLSL code generation
    has been added. Please refer to the dxr example for an illustrative path tracer
    using it.
  • The CUDA/OptiX backend expects some new functions in the user provided renderer runtime
    to allow using resources unknown at compile-time via argument blocks with class compilation:
    • bool tex_texture_isvalid(Texture_handler_base const *self, tct_uint texture_idx)
    • void tex_resolution_3d(int result[3], Texture_handler_base const *self, tct_uint texture_idx)
    • bool df_light_profile_isvalid(Texture_handler_base const *self, tct_uint resource_idx)
    • tct_float df_light_profile_power(Texture_handler_base const *self, tct_uint resource_idx)
    • tct_float df_light_profile_maximum(Texture_handler_base const *self, tct_uint resource_idx)
    • bool df_bsdf_measurement_isvalid(Texture_handler_base const *self, tct_uint resource_idx)
      The tex_resolution_3d() function fills the width, height and depth for the given
      texture index into the respective result entry.
      The other functions are implementations for the corresponding MDL functions.
      See examples/mdl_sdk/shared/texture_support_cuda.h for an example implementation.
  • The compiler support for the ternary operator on material and material sub types has been
    improved. Several materials that caused compile errors before are now compiled flawless.
  • The MDL compiler now correctly issues an error when the called object is not a function,
    detecting (wrong) code like f()().
  • The compiler generated now correct MDL code when exporting functions containing a
    dangling if construct.
  • The compiler now (correctly) forbids the use of resource types as parameter types
    in annotations.
  • The PTX backend does not use global counters to generate temporary identifiers anymore,
    this greatly improves PTX cache hits.
  • The following MDL 1.5 features are now supported by the MDL compiler:
    • hair_bsdf() type
    • df::chiang_hair_bsdf()
    • anno::origin annotation
    • support for annotations on annotation declarations
  • The backends were upgraded to use the LLVM 7 library. This means, that the LLVM-IR
    backend now produces LLVM 7 IR code.
  • Allow generating code for bsdf() if "compile_constants" options is "on" (default).

• MDL SDK examples

  • A new Direct3D 12 example dxr has been added which illustrates how to use
    the HLSL back-end in an RTX-based real-time path tracer.
  • Support for thin-walled materials has been added to example_df_cuda.

Fixed Bugs

• General

  • Failing MDLE creation when an argument of the MDLE prototype is connected to a function
    which returns a user-defined type has been fixed.
  • A bug leading to different output (and therefore different hashes) when exporting the
    same MDLE more than once has been fixed.
  • A failure (error code -8) when creating presets from functions with user-defined
    return types has been fixed.
  • A failure (error code -8) when creating function presets from MDL modules
    with versions < 1.3 has been fixed.
  • When exporting presets from MDLE definitions or MDL definitions containing calls to MDLE
    definitions in their arguments, the MDLE code is now inlined into the new module, rather
    than resulting in invalid MDL.
  • The missing origin annotation on the main definition of an MDLE file has been added.
  • Issues resolving MDLE files on UNC file paths have been fixed.
  • Missing imports for user-defined function return types which caused MDLE creation
    to fail, have been added.
  • The conversion of array constructors to MDL AST expressions has been fixed.
  • The use of implicit conversion functions inside enable-if expressions is no longer
    forbidden.

• MDL Compiler and Backends

  • Unnecessarily slow code generation for very big output has been fixed.
  • Wrong code generation for df::bsdf_measurement_isvalid() has been fixed.
  • Creating DAG call nodes for ternary operators returning non-builtin types has been fixed.
  • For the native and PTX backends, wrong order of array elements returned by math::modf()
    has been fixed.
  • Native code execution on Mac has been fixed.
  • Struct member access for MDLE types containing a dot in their file path has been fixed.

MDL SDK 2019 (314800.830): 20 Feb 2019

ABI compatible with the MDL SDK 2019 (314800.830) binary release
(see https://developer.nvidia.com/mdl-sdk)

Added and Changed Features

• General

  • This release contains a preview version of the new MDL encapsulated (MDLE) file format. Please note that MDLE files generated with this SDK version will be invalidated with the next update.
  • A new API component mi::neuraylib::IMdle_api has been added, which can be used to create MDLE files.
  • The function mi::neuraylib::IMdl_compiler::load_module() can now load MDLE files. Please note that the option bool "experimental" has to be set in the mi::neuraylib::IMdl_execution_context in order to enable support for MDL 1.5, which is needed for MDLE files.
  • The functions
    • mi::neuraylib::IMaterial_instance::is_default() and
    • mi::neuraylib::IFunction_call::is_default() have been added.
  • The standalone tool mdlm has been extended with MDLE specific commands.
  • The class mi::neuraylib::IBaker has been extended to support baking of constants of type mi::IBoolean.

• MDL SDK examples

  • A new example example_mdle has been added to illustrate the use of MDLE files.
  • The example example_df_cuda has been adapted to allow loading and rendering of MDLE files.
  • A new example example_generate_mdl_identifier that illustrates how to generate a valid mdl identifier, e.g., a module name, has been added.
  • The example mdl_browser has been extended to display MDL keywords in the info tooltip as well as above the description in list view mode.

Fixed Bugs

• General

  • A bug when translating light profile and bsdf measurement constructors from the MDL SDK API representation to the MDL Core representation has been fixed.

• MDL Compiler and Backends

  • The hash calculation for struct field access DAG calls for the PTX code cache has been fixed.
  • The handling of array parameters in class compilation has been fixed.
  • A crash when trying to fold an invalid array constructor has been fixed.
  • Missing parentheses when printing operators with the same precedence as MDL has been fixed ("a/(b*c)" was printed as "a/b*c").
  • A potential crash when generating code for distribution functions has been fixed.
  • Wrong error messages "varying call from uniform function" have been fixed, which were generated by the MDL compiler under rare circumstances for struct declarations.
  • Wrong error messages "function preset's return type must be 'uniform T' not 'T'" have been fixed, which were generated by the MDL compiler for function variants if the original function always returns a uniform result but its return type was not declared as uniform T.
  • A discrepancy between code execution on CPU and GPU for constant folding of
    sqrt(c) (c < 0) has been fixed. Now NaN is computed for both.

MDL SDK 2018.1.2 (312200.1281): 11 Dec 2018

ABI compatible with the MDL SDK 2018.1.2 (312200.1281)
binary release (see https://developer.nvidia.com/mdl-sdk)

Added and Changed Features

• MDL 1.5 Language Specification

  • A first pre-release draft of the NVIDIA Material Definition Language 1.5: Appendix E - Internationalization has been added to the documentation set.

• General

  • Support for the internationalization of MDL string annotations has been added. See the MDL 1.5 Language Specification for details.

  • A new API component mi::neuraylib::IMdl_i18n_configuration has been added, which can be used to query and change MDL internationalization settings.

  • A new standalone tool to create XLIFF files has been added. See i18n.

  • Calling mi::neuraylib::ITransaction::remove() on an MDL module will cause the module and all its definitions and other dependencies to be removed from the database as soon as it is no longer referenced by another module, material instance or function call. The actual removal is triggered by calling mi::neuraylib::ITransaction::commit().

  • A new API component mi::neuraylib::Mdl_compatibility_api has been added which allows to test archives and modules for compatibility.

  • A new standalone tool to manage MDL archives has been added. See mdlm.

  • A new API class mi::neuraylib::IMdl_execution_context intended to pass options to and receive messages from the MDL compiler has been added.

  • A new API class mi::neuraylib::IMessage intended to propagate MDL compiler and SDK messages has been added.

  • A new API function mi::neuraylib::IMdl_factory::create_execution_context has been added.

  • The signatures of the API functions

    • mi::neuraylib::IMaterial_instance::create_compiled_material()
    • mi::neuraylib::IMdl_compiler::load_module()
    • mi::neuraylib::IMdl_compiler::load_module_from_string()
    • mi::neuraylib::IMdl_compiler::export_module()
    • mi::neuraylib::IMdl_compiler::export_module_to_string()
    • mi::neuraylib::IMdl_backend::translate_environment()
    • mi::neuraylib::IMdl_backend::translate_material_expression()
    • mi::neuraylib::IMdl_backend::translate_material_df()
    • mi::neuraylib::IMdl_backend::translate_material()
    • mi::neuraylib::IMdl_backend::create_link_unit()
    • mi::neuraylib::IMdl_backend::translate_link_unit()
    • mi::neuraylib::ILink_unit::add_environment()
    • mi::neuraylib::ILink_unit::add_material_expression()
    • mi::neuraylib::ILink_unit::add_material_df()
    • mi::neuraylib::ILink_unit::add_material()

    have been changed to use the new class mi::neuraylib::IMdl_execution_context.
    The old versions have been deprecated and prefixed with deprecated_. They can
    be restored to their original names by setting the preprocessor define
    MI_NEURAYLIB_DEPRECATED_9_1.

  • The API functions

    • mi::neuraylib::IMdl_backend::translate_material_expression_uniform_state()
    • mi::neuraylib::IMdl_backend::translate_material_expressions()

    have been deprecated and prefixed with deprecated_. They can be restored to
    their original names by setting the preprocessor define MI_NEURAYLIB_DEPRECATED_9_1.

  • The utility classes

    • mi::neuraylib::Definition_wrapper and
    • mi::neuraylib::Argument_editor

    have been extended to provide member access functions.

• MDL Compiler and Backends

  • Support for automatic derivatives for 2D texture lookups has been added to the PTX,
    Native x86 and LLVM IR backends. This feature can be enabled via the new backend option "texture_runtime_with_derivs". Please refer to the "Example for Texture Filtering with Automatic Derivatives" documentation for more details.
  • Measured EDFs and BSDFs can now be translated to PTX, Native x86 and LLVM IR. Note that the texture runtime needs to be extended with utility functions that enable runtime access to the data.
  • Spot EDFs can now be translated to PTX, Native x86 and LLVM IR.
  • The nvidia::df module has been removed.

• MDL SDK examples

  • Support for automatic derivatives has been added to the example_execution_native,example_execution_cuda and example_df_cuda examples, which can be enabled via a command line option.
  • The example_execution_native example has been extended to allow to specify materials on the command line. It is now also possible to enable the user-defined texture runtime via a command line switch.
  • The CUDA example texture runtime has been extended with support for measured EDF and BSDF data.
  • The MDL Browser is now available as a QT QML Module which can also be integrated in non-qt based applications.
  • Initial support for class compiled parameters of type Texture, Light_profile, and Bsdf_measurement has been added to example_df_cuda. So far, it is only possible to switch between all loaded resources, new resources cannot be added.

Fixed Bugs

• General

  • An error when exporting presets where MDL definitions used in the arguments require a different version than the prototype definition has been fixed.

• MDL Compiler and Backends

  • A missing check for validity of refracted directions has been added to the generated code for the evaluation of microfacet BSDFs.
  • Incorrect code generation for math::length() with the atomic types float and double has been fixed.
  • The computation of the minimum correction pattern in the MDL compiler has been fixed.
  • The compilation of || and && inside DAG IR has been fixed.
  • Pre and post increment/decrement operators when inlined into DAG-IR have been fixed.
  • Previously missing mixed vector/atomic versions of math::min() and math::max()
    have been added.
  • The handling of (wrong) function references inside array constructor and init constructor has been fixed, producing better MDL compiler error messages.
  • The hash computation of lambda functions with parameters has been fixed.
  • If an absolute file url is given for a module to be resolved AND this module exists in the module cache, the module cache is used to determine its file name. This can speed up file resolution and allows the creation of presets even if the original module is not in the module path anymore.
  • A memory leak in the JIT backend has been fixed.
  • The generated names of passed expressions for code generation have been fixed.

Known Restrictions

• When generating code for distribution functions, the parameter global_distribution on spot and measured EDF's is currently ignored and assumed to be false.

MDL SDK 2018.1.1 (307800.2890): 15 Sep 2018

ABI compatible with the MDL SDK 2018.1.1 (307800.2890)
binary release (see https://developer.nvidia.com/mdl-sdk)

Added and Changed Features

• General

  • A new API function mi::neuraylib::ILink_unit::add_material has been added to
    translate multiple distribution functions and expressions of a material at once.
  • A new API function mi::neuraylib::IMdl_backend::translate_material has been added to
    translate multiple distribution functions and expressions of a material at once.
  • A new function mi::neuraylib::ICompiled_material::get_connected_function_db_name
    has been added.
  • A new function IImage_api::create_mipmaps has been added.
  • A new parameter has been added to the functions mi::neuraylib::ITarget_code::execute*
    to allow passing in user-defined texture access functions.

• MDL Compiler and Backends

  • Diffuse EDFs can now be translated to PTX, native x86 and LLVM IR.
  • Support for passing custom texture access functions has been added to the Native backend.
    The built-in texture handler can be disabled via the new backend option
    "use_builtin_resource_handler".

• MDL SDK examples

  • The example_df_cuda example now features simple path tracing inside the sphere to
    enable rendering of transmitting BSDFs.
  • To allow loading of multiple materials within a module, a wildcard suffix "*" is now
    supported in the material name command line parameter of the example_df_cuda example.
  • The example_df_cuda has been updated to illustrate the use of the new function
    mi::neuraylib::ILink_unit::add_material.
  • The example_execution_native has been extended to illustrate the use of user-defined
    texture access functions.
  • The example_mdl_browser can now be built on Mac OS.

Fixed Bugs

• General

  • The handling of archives containing a single module has been fixed in the
    mi::neuraylib::IMdl_discovery_api.
  • The handling of relative search paths has been fixed in the
    mi::neuraylib::IMdl_discovery_api.

• MDL Compiler and Backends

  • Various fixes have been applied to the code generated for BSDF's:
    • The computation of the evaluate() function for glossy refraction has been fixed
      (df::simple_glossy_bsdf, df::microfacet*).
    • The sample() functions for layering and mixing now properly compute the full PDF
      including the non-selected components.
    • The implementation of df::color_clamped_mix() has been fixed
      (the PDF was incorrect and BSDFs potentially got skipped).
    • All mixers now properly clamp weights to 0..1.
    • Total internal reflection is now discarded for glossy BSDF
      (df::simple_glossy_bsdf, df::microfacet*) with mode df::scatter_transmit,
      as defined in the MDL specification.
  • Incorrect code generation for math::normalize() with the atomic types float and
    double has been fixed.
  • The generation of function names for array index functions for modules in packages
    has been fixed.
  • In rare cases, compilation of a df* function could result in undeclared parameter names
    (missing _param_X error), which has been fixed.
  • The compilation of MDL presets of re-exported materials has been fixed.
  • In rare cases, the original name of a preset was not computed, which has been fixed.

MDL SDK 2018.1 (307800.1800): 09 Aug 2018

• Initial open source release
• ABI compatible with the MDL SDK 2018.1 (307800.1800) binary release (see https://developer.nvidia.com/mdl-sdk)
• The following features are only available in the binary release and excluded in the
  source code release:
  • MDL distiller
  • Texture baking (see examples/mdl_sdk/execution_cuda for example code for texture baking)
  • GLSL compiler back end
• Added: MDL Core API, a lower-level compiler API in the MDL SDK (see src/prod/lib/mdl_core and doc/mdl_coreapi)
• Added: examples for the MDL Core API (see examples/mdl_core and doc/mdl_coreapi)