mental ray Release Notes

Version 3.7.53.5

May 18, 2009

Introduction
Changes by Version
Support for OpenEXR

Introduction

These release notes list changes in mental ray since version 3.6.0.3.

Changes by Version

Note: Version 3.7+ and later versions of mental ray are binary incompatible with mental ray versions 3.6/3.6+ and prior. Partial compatibility with 3.7 is maintained.

Some shaders compiled for previous versions may need recompilation to be used with this version. Shaders compiled for mental ray 3.7 will work with mental ray 3.7+, but not the other way round. This specifically applies for the MetaSL generated shaders.
Version 3.7+ cannot be combined with version 3.7 and earlier versions in network rendering.

Changes in version 3.7.53.5

Bug fixes:

Fixed possible crash in ccmesh object tessellation.
Fixed a crash if progressive rendering is used in combination with contours: disabled contour rendering. These two features are algorithmically incompatible.

Changes in version 3.7.53.4

Bug fixes:

For finalgather precomputation, the face flag for the eye ray has been mistakenly initialized from the options block in a way, so that kernel considered it as set by shader and gave it higher priority than the object / instance flags, which thus have been ignored.
For possible crash or database fatal for scenes which do not fit into memory completely. A reference to subobject could stay in the kernel after the complete object with subobjects was flushed for the second time.

Changes in version 3.7.53.3

Shader changes:

For architectural material shader, changes the installed callback calls to unding the sample loop osurround the sample loop op of the light, instead of surrounding the light loop. Adding light data to the callback argument structure.

Changes in version 3.7.53.2

Bug fixes:

For BSP2 acceleration used with shutter_delay = shutter > 0, fixed possible shadow artifacts due to the static BSP2 tree used in this case. Note that with this fix, the memory allocated by the BSP2 acceleration is increased, but stays below the memory required for motion BSP2 acceleration.
Fixes possible hangs and z-clipping artifacts in rendering of scenes with motion blur with rasterizer. The precision of the z-clipping computation is increased.
For the debug version executable, fixed false assert in finalgather.

Shader changes:

For architectural material shader, fixed a performance regression in case of multibouce finalgathering used in combination with "fake glossy reflection" feature of the shader.

Changes in version 3.7.53.1

Feature improvements:

For progressive rendering, added color profiles support for interactive image output. Optimized pipeline for color profile image tile conversion. In the case of progressive rendering, apply fast but slightly less accurate conversion routines. Speedup general image conversion routine by avoiding unnecessary function calls and memory allocations.
For progressive rendering, adjusted the set of exposed string options and command line commands:
-progressive <bool>

-progressive_subsampling_size <int>

-progressive_min_samples <int>

-progressive_max_samples <int>

-progressive_error_threshold <float>

Corresponding string options are "progressive", "progressive subsampling size", "progressive min samples", "progressive max samples" and "progressive error threshold".

The -progressive_subsampling command line option has been renamed to match the string option name.

The new options set allows usage of progressive rendering with adaptive sampling. The "progressive min samples" specifies the minimal number of samples used. "progressive max samples" and "progressive error threshold" specify the maximal number of samples and estimated error level to be used per pixel before stop, whatever criteria is reached first.

Bug fixes:

For multiple bounce finalgathering, fixed wrong depth count for secondary finalgather rays. Refraction limit has been applied instead of reflections.
For multiple bounce finalgathering, fixed wrong default importance setting applied to the 3rd and higher diffuse bounces. This resulted in a higher number of rays used, shifting the scene rendering towards higher quality on the same settings.
Path substitution rules have not been applied to the file names of framebuffers. One of the effects of this bug was that the small stub file for imf_disp interactive display has been created at the correct location, but the framebuffer file has been saved somewhere else or failed to save. Fixed.
On Windows, fixed initialization of the noise table with changing values for several mental ray invocations.
For multihosted rendering, fixed possible crash at the end of the frame for scenes using assemblies. Some scenes with BSP2 acceleration may be affected as well.
Fixed potential rare crash if mental ray was low on memory and flushed parts of the scene.
For integrated version of mental ray, fixed possible rendered tile tag/memory leak in case of aborting rendering.

Shader changes:

For architectural material shader, bump vector with zero first 2 component but not-zero 3rd component has not been applied, fixed.

Changes in version 3.7.53.0

Feature improvements:

For the scene echo, if the value of an miBoolean is neither miTRUE nor miFALSE, echo it as an integer.
Improved reaction time on rendering aborts, especially those during finalgather precomputation stage displayed interactively. mental ray kernel copies rendered tiles to the framebuffer asynchronously. The outstanding tiles are no longer copied after abort signal.
For progressive rendering, intercept invalid subsample size values below 1.
For MetaSL shaders, added support for light loops light_dot_nl, light_uv, light_area and light_area_delta state variables.
For MetaSL shaders, initialize light_distance_limit with huge scalar.
For MetaSL shaders, renamed some runtime header files.
For MetaSL shader runtime, improve robustness of the tangent space computation.

Bug fixes:

Fixed a major memory leak regression in case of 8bit rendered images in combination with virtual framebuffers (-fb_virtual on) and interactive display (imf_disp) attached.
For hair rendering, fixed possible artifacts due to the hair radius not taken into account when bounding boxes were computed.
For ray-traced hair with motion blur, fixed possible rendering artifacts due to too small bounding boxes used.
Fixed rendering artifacts for objects with number of motion segments higher than the global motion steps options.
For scenes with assemblies, fixed possible freeze and memory leak on assembly deletion at the end of the frame rendering.
For ambient occlusion, fixed possible artifacts and assertion in the debug version if a negative number of AO rays was requested or if a negative importance factor has been used.
For multihosted rendering used in heterogenous network, fixed possible crashes when rendering scenes with finalgathering. The crash was due to binary incompatibility of the network protocol for 32 and 64-bit machines.
For multihosted rendering, fixed possible rare crash due to inappropriate pincounting.
Fixed possible crash at shutdown if cached framebuffers (command line -fb_virtual cache) were used.
For mental ray used with cached framebuffers (command line -fb_virtual cache) fixed possible short delays when multiple tiles appears in display window in bunches. The tiles displaying is made more uniformly in time.
Fixed a regression in MetaSL compiler causing a crash during the internal generation of shader declaration.
For MetaSL shaders, added missing support for the light_type member of the light state
For MetaSL compiler, fixed the order of handling for include files and user types declared in such files.
For MetaSL compiler, fixed generation of the code for matrix operator*= operators.
For MetaSL compiler, fixed compilation of Type t(); and Type t=Type(); constructor expressions.
For MetaSL shaders, fixed compilation of a shaders returning structs containing substructs.
For MetaSL CIL backend, fixed compilation of some matrix constructors and 4-vectors constructed from a couple of 2-vectors.
For MetaSL CIL backend fixed compilation of shaders with int2, int3, bool2 and bool3 parameters.
For MetaSL shader integration, fixed possible memory corruption for shaders and phenomena using textures and other parameters identified by name.
For MetaSL C++ backend, properly initialize array fields of structs in the structs default constructor. This fixes a runtime crash if such constructors were used.
For MetaSL C++ backend, fixed illegal code generation: generate user-defined types before functions defined in #include files.
For MetaSL C++ backend, fixed possible illegal code generated for right hand side swizzles.

Shader changes:

For base and subsurface scattering lightmap shaders, significantly reduced memory consumption for scenes with large triangles covering high number of pixels in the lightmap mesh rendering.
Fixed missing output passes in the "mi_car_paint_phen_x" phenomenon.

Changes in version 3.7.52.15

Bug fixes:

Fixed conversion of scalar half-float textures.
Added missing support for half-floats for pass files.

Changes in version 3.7.52.14

Feature improvements:

On Windows platforms, avoid popup windows for integrated version of mental ray library when external commands for shader compilation are invoked. Affected are MetaSL shaders with CIL and C++ backend, as well as shaders provided in form of C/C++ source code, as .obj object file and as .lib static libraries.
When such external compiler, linker, CIL assembler or manifest embedding is invoked, its output to stdout is captured by mental ray kernel and printed out as LINK module debug messages. The output to stderr is printed as LINK module error messages. Some of the external tools put error-style messages into stdout, so for troubleshooting it is recommended to run mental ray with debug output (-v 6 command line option for mental ray standalone).

Bug fixes:

Fixed a crash if a very long (8 kilobytes) error/warning/info/debug message has been created. On Windows platforms, a protection mechanism for such messages issued from shaders existed prior this version.
When parsing a .mi file, allow includes of .msl / xmsl files between top-level grammar elements only. This ensures that elements with no grammatic "end" marker, like textures, are completed before the included file is processed.
Fixes the issue on texture declarations described in the release note for version 3.7.52.13.
For MetaSL shaders, fixed several issues in initialization of arrays. This fixes possible crash at runtime when array elements were used.
For MetaSL compiler, make sure that string initializers for textures and light profiles are not accepted for local variables.
For MetaSL compiler, fixed lerp function compilation.
For MetaSL compiler, removed left-over code dumping ambiguous overloads.
For MetaSL CIL backend, fixed code generation for constructors of structs.
For MetaSL CIL backend, fixed code generation for functions with array parameters.
For MetaSL CIL backend, fixed several minor memory leaks.

Shader changes:

For MetaSL noise generation shaders, replaced currently not supported swizzles in the parameter initializers with equivalent code without swizzles.

Changes in version 3.7.52.13

New features:

Standalone version of mental ray can be used as a MetaSL compiler directly. Command line options following the new -mslc option are passed to the compiler. The compiler is invoked instead of mental ray renderer if the option is present. For example, the combination -mslc -h will print out compiler options.
Current version of the compiler can be used to produce C++ (-t cpp) and .mi declaration (-t mi) files, hardware shader compilation is not included.

Feature improvements:

In mental ray options/camera info banner printed before rendering, use semantics / name of the framebuffer instead of the framebuffer index.
Improved performance of the DISP module in case imf_disp is attached to an image rendered with 16bit, float or half pixels. The improvement is primarily relevant for progressive rendering. This improvement does not apply to cached (tiled) framebuffers, which are not recommended for usage with progressive rendering for general performance reasons.
For MetaSL C++ backend, implemented support for texture resources specified as default MetaSL shader input parameter values.
Note that such textures need to be declared before including XMSL files, like

filter 1.0 color texture "default.dds" "MyDefaultTexture.jpg"

Due to the way mental ray parser works in the current version, it is also required to explicitly terminate texture grammar before the including msl/xmls files. This can be done with a "set" statement like:

set "MetaSL default textures provided" "on"

The latter limitation will be removed in the future versions (fixed in 3.7.52.14).
Output messages from MetaSL compiler are now given according to compiler submodules, not "LIB" as in the previous versions.
For MetaSL C++ backend, generate framebuffer access code only when required. This removes frequent C++ compiler warning on unused fb_index variable.

Bug fixes:

For mental ray 3.5 framebuffer compatibility interface, fixed wrong handling of user framebuffers 6 and 7.
For the second "talk" version of the DISP interactive display protocol, fixed wrong image pixel depth. For the primary color buffer, the actual pixel type used in the framebuffer is announced and sent through the protocol.
If the primary framebuffer is known under multiple names, all names are listed with the fb_list_name command.

Fixed possible wrong indices for some framebuffers.
For progressive rendering, fixed possible crash if rendering in a low image resolution.
For progressive rendering, fixed a crash if output shaders were used.
For standalone mental ray, fixed a crash if more than one -texture_path options was used on the command line.
For recently added internal support for half type, added missing clipping in mi_img_clip_color function.
For MetaSL compiler, fixed a crash if both input parameter initializer and annotations were present.
For MetaSL compiler, added missing constructors for Trace_options.
For MetaSL compiler, fixed possible crash in nested switch operator.
For MetaSL compiler and backends, added missing is_nan and is_finite functions.
For MetaSL compiler, fixed conversions between float3, float4, Color and Spectrum types.
For MetaSL C++ backend, fixed possible broken generation of the code for texture space access.
For MetaSL CIL backend, fixed compilation failures for shaders using ray types and light types.
For MetaSL CIL backend, fixed compilation of shaders using clamp function.
For MetaSL CIL backend, fixed compilation of shaders using any and all functions.
For MetaSL CIL backend, fixed compilation of shaders using abs function on integers and integer vectors.
For MetaSL CIL backend, fixed string comparison operators.
For MetaSL CIL backend, added support for member constructors and destructors.
For MetaSL CIL backend, fixed a regression for shaders using ray types.
For MetaSL shaders runtime, avoid asserts in creating local orthonormal basis
For MetaSL shaders runtime, request to unsupported value of type float2 returned uninitialized value, added initialization with zeroes.

Shader changes:

Fixed MetaSL shader source code for shaders using reserved keywords.

Known MetaSL limitations in this version:

The following parts of MetaSL 1.0 specification are not supported by mental ray 3.7+, and not planned to:
- Material Phenomena
- Techniques
- 3D and Cube textures
- MetaSL Spectrum is mapped onto mental ray color type
- Particle maps are not supported
- Some limitations on array and struct input/outputs apply. In particular, output array parameters are not supported.
- Native functions
- Explicit shader invocation
- Derivatives and ray differentials
mental ray shader parameter type is a subset of MetaSL type system. A parameter type conversion mechanisms is applied which may result in precision loss.
Textures for default texture parameters need to be declared explicitly before MetaSL shaders.
MetaSL CIL backend has worse performance compared to C++ backend. C++ backend is the recommended solution.
The following issues are known in this versions but planned to be fixed in the mental ray 3.7+ final release:
- For input parameters of MetaSL shaders, initializers using swizzles are not supported.
- Texture blur is not supported.
- Some of the light state members are not supported.
- CIL backend is less complete than C++ backend. For CIL backend, default input texture parameters are ignored. Some of initializers are not working properly.
- Switch between (precompiled) C++ shader backend and CIL backend is global, not per shader.

Changes in version 3.7.52.12

Feature improvements:

Implemented detection if a shader parameter has been assigned a value. This mechanism is required in order to allow MetaSL shader input parameter constructors which are not constant values. The order of default values is:
- .mi file or mental ray API value of a shader function.
- Default value of the input parameter in the MetaSL code.
- Default value of the input parameter in .mi file or mental ray API shader declaration.
- Zero value / black.
In order to accommodate this changes, the i_is_connected mi_is_connected macro has been added. The miFunction::has_connection field has been added, which is set if at least one shader parameter is connected.

Note that this change may affect shader binary compatibility:

Old shaders using mi_eval and mi_call macro for parameter evaluation will work with the new kernel without changes and recompilation. For the performance reasons it is recommended to recompile these shaders with the new shader.h header file.

Shaders which accessed internal ghost tags directly in order to check for the connected parameters may require minor source code modification. The mi_is_connected macro should be used instead of direct access.
Implemented support for "half" (16-bit floats). Prior versions of mental ray supported 16-bit-float OpenEXR textures, but the internal representation used 32-bits. Native 16 bit representations saves almost 50% of memory occupied by textures.
Note that old .map textures of rgb_h/rgba_h image types are binary incompatible with the new version of mental ray. If used, such textures will be loaded into memory and converted, but they will not be memory-mapped.
For MetaSL C++ backend, implemented support for non-constant input parameter initializers in MetaSL shaders.
Reduced memory consumption of the MetaSL CIL backend code generator and improved its performance.

Bug fixes:

If compatibility framebuffer interface was used, user framebuffers 6 and 7 have not been written out correctly, fixed.
For MetaSL compiler, add a warning for older MetaSL phenomena assigning values to output parameters in XMSL files. These values are ignored.
For MetaSL compiler, changes the names of texture samplers according to the MetaSL specification.
For MetaSL compiler, added missing implicit type conversion handling for the conditional "? :" expressions.
For MetaSL compiler, fixed a regression in compilation of abs, acos and some other builtin functions.
For MetaSL compiler, fixed comparisons operators for textures and sampler.
For MetaSL compiler, avoid assigning values directly to the xmsl phenomena output parameters.
For MetaSL compiler, fixed operator %=.
For MetaSL compiler, fixed vector *= matrix and matrix *= vector operators.
For MetaSL compiler C++ backend, fixed possible crash when arrays of length zero were used.

Shader changes:

For the mip_render_subset, fixed possible artifacts if rendering of placeholder objects in combination with coverage framebuffer filtering was used.
For the production mip_motionblur shader, fixed possible black stripe artifacts.
For the paint shaders, the mi_metallic_paint_output_mixer was not linked into the library, fixed.
For the architectural material shader, fixed possible NaN artifacts originating from dot normal precision.

Changes in version 3.7.52.10

Feature improvements:

Extended the second "talk" version of the DISP interactive display protocol with the "fb_list_name" command. It lists currently active framebuffers, one line per buffer:
fb_list_name: <framebuffer index>,<int image type> <image type name with +-prefix, like +rgba>, <int resolution_x> <int resolution_y>, <int number of image components> <bits per pixel component>, "<framebuffer_name>"

The framebuffer name is quoted. Also fixed a bug in fb_list command which did not list all the framebuffers.the second version,

Bug fixes:

For progressive rendering, fixed a possible crash in displaying miscounted DISP tiles.
For MetaSL shaders compiled with flattened enabled, fixed missing compilation of light shaders.
For MetaSL compiler, added missing implicit type conversion handling for the conditional "? :" expressions.
For MetaSL C++ backend, avoid a compiler crash if no main method is present in a shader source.
For MetaSL C++ backend, added missing support for shader static member constructors.
For MetaSL CIL backend, fixed wrong code generation for switch used with non-consecutive enum case labels.
For MetaSL CIL backend, fixed generation of the code for construction of matrices from matricide of some other dimensions.
For MetaSL CIL backend, fixed generation of the code with changed capitalization of texture sampler methods, follows recent MetaSL specification adjustment.

Shader changes:

For the mip_metallic_paint(_x) shader, fixed wrong illumination if multiple area light sources are present in the scene.
For the production mip_rayswitch_stage shader, fixed declaration apply mask.

Changes in version 3.7.52.9.

Bug fixes:

Fixed a crash for MetaSL shaders using Spectrum type.
For MetaSL shaders, added support for input initializers with direct state variable assignments (like input: a = normal;). Note that in this version such initializers may overwrite the explicit (not connected) value for parameters provided by mental ray API / .mi syntax.
For MetaSL shaders, added support for default-to-state annotations. Note that this mechanism is deprecated, support for shader parameter initializers will follow.
For MetaSL CIL backend, fixed possible running out of memory crash if a large number of shaders need to be compiled.
For MetaSL CIL backend, fixed broken code generated for matrix constructors.
For MetaSL compiler, fixed broken code produced for struct input parameters containing textures.

Changes in version 3.7.52.8

Bug fixes:

For multihosted rendering, fixed possible odd warning on unfreezing camera.

Changes in version 3.7.52.7

Feature improvements:

For Image Based Lighting (IBL), added new string option "environment lighting shadow". Possible values are:
- "off": no visibility tests, no shadows from the environment,
- "solid": all objects (including transparent/semi-transparent ones) cast solid shadows,
- "transparent": support for transparent and semi-transparent objects.
The default value is "transparent". Earlier versions of mental ray follow the "solid" option. Rendering with "solid" option is slightly faster.
For MetaSL shaders, fmod function for integers has been deprecated. Replaced it with operator%.
The default value of the registry {_MI_REG_METASL_EMULATE_MI_PHEN} introduced in version 3.7.52.0 to convert MetaSL phenomena into mental ray phenomena for usage with precompiled shaders has been changed to false.

Bug fixes:

For builtin architectural material BSDF, fixed too low reflections under the Fresnel layer.
For MetaSL shaders, fixed usage of info,warning,error functions in free (non-shader) functions.
For MetaSL shader light loops, the get_light_parameter() function returned true even if the parameter did not exist. Fixed.
For MetaSL shade light loops, pass light direction properly.
For MetaSL shaders, fixed a regression in shadow shaders: the transparency value was taken as a result instead of applying it as multiplier.
For precompiled MetaSL shaders, fixed generation of shader declarations with no inputs.
For precompiled MetaSL shaders, fixed generation of shader declarations with inputs in "float a, b;" style.
For MetaSL shaders, fixed compilation of implicit texture to boolean conversion.
For MetaSL CIL shaders, fixed wrong code generated for assignments to vector elements selected by indexing.
For MetaSL shaders compiled to C++, fixed possible broken generated C++ code in array initialization.
For MetaSL compiler, fixed possible crash when an input parameter was declared as an unsized array with no initializer.
Fixed some memory leaks in MetaSL compiler.

Changes in version 3.7.52.5

Feature improvements:

Improved error reporting for MetaSL phenomena.

Bug fixes:

For multi-hosted rendering of scenes with BSP2 accelerations, fixed possible crash in rendering of second and subsequent frames.
On Windows platforms: for multi-hosted rendering of scenes with BSP2 acceleration, fixed possible freeze in rendering of second and subsequent frames.
Fixed a crash if photon map diagnostic was used in combination with -photonmaps only command line option or the corresponding rendering mode for integrated versions.
For precompiled MetaSL shaders, fixed normal connection for bump shaders.
For MetaSL shaders running in precompiled mode, fixed a crash for phenomena having multiple outputs.

Changes in version 3.7.52.4

Feature improvements:

Improved performance and scalability of progressive rendering.
For MetaSL shaders, improved support for ray type.

Bug fixes:

For ccmesh objects, improved possible artifacts originated from insufficient numerical prevision of derivatives.
Fixed double contribution of the environment lighting if finalgather and IBL has been enabled.
For progressive rendering, fixed wrong broken handling of events.
For MetaSL BSDF shaders, avoid shooting unneeded rays in the rendering kernel.
For MetaSL compiler, fixed several bugs in MetaSL phenomena, including crashes of loading of some .xmsl files.
Fixed initialization of arrays inside of MetaSL phenomena. This affects MetaSL internal nodes only, mental ray shader parameters do not support initializers.

Changes in version 3.7.52.3

Feature improvements:

For high resolution images rendered with finalgather, substantially reduced the amount of continuous memory needed by multi-pass finalgather precomputation. This allows for bigger images to be rendered if memory is fragmented.
For MetaSL shaders, MetaSL shader parameters of type float2 are now mapped to mental ray API parameter type miVector. float4 maps to miColor. int2/3/4 and bool2/3/4 map to miVector (for 2 and 3 values) and miColor (4 values), respectively. MetaSL matrices of all types are mapped to miMatrix.
In the MetaSL CIL backend, a number of missing MetaSL features have been implemented. This includes (but is not limited to):
- assignment to state variables
- fetching multiple outputs of a shader in a trace call
- texture members
- texture samplers
- ddx and ddy
Beyond this various bugs have been fixed, most notably in the code generated for swizzles, the mapping of some types and operators, and in the handling of implicit conversions.
For precompiled MetaSL shaders, shading graph connections to normals and positions are now supported to allow for effects such as bump mapping and normal mapping.

Bug fixes:

For multihosted rendering, fixed memory leaks on slave hosts that occurred after frame 1 in case multiple frames were rendered. When BSP acceleration was used, these leaks were significant.
For multihosted rendering, fixed distribution of MetaSL CIL shaders. In order to use the CIL backend, all slaves hosts must run under Windows.
In the MetaSL compiler, fixed a possible memory corruption which occurred in multihosted network rendering, or debug versions of the compiler.

Changes in version 3.7.52.2.

Feature improvements:

For MetaSL shaders, added support for Lambert BSDF.
For the API consistency reasons, renamed classes called Type_access resp. Type_edit to Access_type resp. Edit_type. Corresponding typedefs are provided for compatibility.
In debug version of mental ray: for MetaSL CIL backend, generate debugging information.

Bug fixes:

Fixed possible shadow artifacts in Image Based Lighting (IBL).
For MetaSL compiler, fixed missing arguments casts in constructors, in particular Color(0, 0, 0, 1) is fixed.
For MetaSL compiler, implemented missing % (mod) operator.
For MetaSL shades the glossy component for direct_light function has been computed without testing the occluders of the environment, fixed.
For MetaSL CIL backend, fixed texture samplers with filtering. Fixed several further issues.
Platform specific: on Linux platforms, no longer link libgcc statically. This has caused crashes in exception handling of MetaSL compiler for some cases then Linux distribution where mental ray has been compiled and executed were different.

Changes in version 3.7.52.0.

Feature improvements:

Introduce C++ API for Catmull-Clark mesh subdivision surface geometry for access and manipulation from shaders. The interface is provided in the new header file shader_sds34.h which is automatically included by shader.h when using C++ compilers.
Reduce peak memory consumption for finalgather progressive preview.
Performance improvement in the progressive rendering mode to achieve a desired quality in less time.
Add motion blur to be rendered in progressive mode.
For MetaSL shader graphs and phenomena provided in .xmsl files, added support for constructing the actual phenomenon in mental ray instead of compiling it into a single mental ray shader. This allows to support MetaSL phenomena that are built externally (like in mental mill) but rendering with shader nodes which have been pre-compiled (from native or MetaSL source code) and pre-loaded into mental ray. This feature explicitly avoids runtime compilation of MetaSL shaders or phenomena. Note, that this behavior is enabled by default. It can be disabled defining the registry key {_MI_REG_METASL_EMULATE_MI_PHEN} and setting its value to off.
For MetaSL shaders, implement support for conditionals on scalars and vector values expressions.
For MetaSL shaders, implemented Texture_info class.
For MetaSL CIL backend, implement support for shadowing from MetaSL light shaders.

Bug fixes:

Fix a bug leading to crashes when using multiple finalgather map files.
For MetaSL CIL backend, fix problems when accessing files in temporary directories with a name that contains spaces.
For MetaSL support, fix recognition of registry keys together with their actual values.
For MetaSL CIL backend, fix more bugs in code generation on both 32bit and 64bit reducing the potential for CIL compilation errors or .NET runtime crashes.

Shader changes:

For architectural material shaders, avoid artifacts when using anisotropic effect in extreme conditions by raising accuracy of computations.
For architectural material shaders, fix uninitialized output values leading to false color artifacts in certain cases.
For production shaders, fix display order of finalgather shooters to be top-left-to-bottom-right.

Changes in version 3.7.51.0.

Feature improvements:

Add new command line option -progress_frequency n and remove former support for an optional second argument to -verbose. This option allows to control the frequency of percentage progressive messages. For example, -verbose 5 -progress_frequency 10 command line would result in printing progress messages for approximately each 10 percent.
On Windows platforms, remove temporary cached framebuffer files is ray executable is killed.
For MetaSL shaders, implemented support for BSDFs (BRDFs).
For MetaSL shaders, implement proper support for ray types in trace functions.
For MetaSL shaders, improve performance for frequent access to shader states.
For MetaSL CIL backend, implement support for unary operators.
For MetaSL CIL backend, implement support for various texture lookup functions.
For MetaSL CIL backend, implement support for missing transformation functions.
For MetaSL shaders, re-enabled message passing across trace() and other functions, which has been temporarily disabled in version 3.7.50.10.
For MetaSL shaders, enabled message passing for light loops.

Bug fixes:

Reduced the memory consumption for final gather progressive preview.
Fixed return value of query miQ_NUM_TEXTURES for hair objects with placeholders.
Fixed possible crash if a phenomenon evaluation was attached to a BSDF slot. In particular, this could be the case when BSDFs were used in combination with Irradiance Particles.
For MetaSL shaders, fixed tracing a ray done instead of environment lookup if trace function was called with environment ray type.
For MetaSL shaders, fixed shader compilation error if variable length array parameter was used.
For MetaSL shaders used as shadow shaders, fixed a crash in MetaSL shader with "transparency" parameter of type float was used.
For MetaSL CIL backend, fix various bugs in code generation reducing the potential for CIL compilation errors or .NET runtime crashes.
For MetaSL CIL backend, fixed shader compilation if temporary directory file name contained spaces.

Changes in version 3.7.50.15.

Feature improvements:

In the debug executables on Windows, changes the implementation of miASSERT to allow to continue execution past the assert point.
For the progressive finalgather precomputation of images with large resolution, significantly reduced the amount of memory used for temporary structures.
For OpenEXR files, changed channel names used to save normal and motion vector framebuffers to NX/NY/NZ and MX/MY/MZ. Previously used name "Y" did not follow the recommendation and confused many viewers for being intensity.
Improved performance of the Quasi-Monte-Carlo sampler generator. The sample patterns are not changed.
Known issue in this version: photon emission from some type of lights may fail leading to "no photons emitted" error (fixed in 3.7.50.16).
For MetaSL shaders, improve error handling by printing symbolic names instead of numeric ids for unsupported state/camera/frame variables. If unsupported normal derivatives are accessed, return zero as approximation instead of printing warnings.

Bug fixes:

For the BSP2 acceleration used in combination with motion blur with multi-segment moving geometry, fixed possible artifacts due to improper computation of the motion segment time interval.
For non-rasterizer rendering modes, clean up the opacity used by mi_opacity_get function for each sample.
For MetaSL shaders, support enabling/disabling environment properly.
For MetaSL CIL backend on 64-bit Windows, generate 64-bit shader libraries instead of 32-bit ones.
For MetaSL CIL backend, fixed light iterators.
For MetaSL CIL backend, implemented missing array support.

Shader changes:

For architectural, paint, production, subsurface and base shaders, miState::importance is now used consistently instead of the miIM key for mi_shaderstate_set/mi_shaderstate_get functions. This provides interoperability with MetaSL shaders and improves performance slightly.
Improved performance of the mip_matteshadow shader by avoiding unnecessary mi_flush_cache calls.
For the mip_matteshadow shader fixed possible finalgather light leak artifacts by using an individual id in order to restrict the finalgather map lookup ( miIrrad_options::id has been introduced in version 3.7.50.4).

Changes in version 3.7.50.14.

Feature improvements:

For the Image Based Lighting (IBL) introduced in version 3.7.50.11, allow to access the build-in environment light by the name "builtin_light_env_inst". This allows to use IBL in the explicit light lists.

Bug fixes:

Fixed possible crash if miRC_TO_SPIRAL rendering tile order was used. Regression in version 3.7.50.11.
For MetaSL CIL backend, fixed enumerations and structs.

Changes in version 3.7.50.13.

Feature improvements:

For lens shaders used with the rasterizer, initialize miState::point, miState::dist and miState::dir according to the closest hit the closest hit.

Bug fixes:

For MetaSL CIL backend, fixed crash in compilation of shaders with no input parameters.
For MetaSL CIL backend, fixed LHS (left hand side) swizzles.

Changes in version 3.7.50.12.

Known limitation is this version: MetaSL CIL backend can crash for shaders with no input parameters, fixed in 3.7.50.13.

Feature improvements:

Changed the license scheme for multi-core CPUs. The limit of 2 or 4 cores per CPU is removed.

Bug fixes:

Fixed a crash in the integrated versions of mental ray using mental ray 3.5 compatibility framebuffer interface. Regression in version 3.7.50.9.
Fixed possible infinite recursion in the error handline of broken PNG images.
For MetaSL CIL backend fixed static functions, fixed component selection and swizzling on vectors.

Changes in version 3.7.50.11.

A temporary CIL backend usage example has been added to common/doc/tutorials/metasl_cil_backend.

New features:

Built-in Image Based Lighting (IBL).
Though image base lighting is implemented in some of OEM shader packages, built-in solution is more efficient and does not require efforts to be invested atop of mental ray kernel to develop corresponding shaders.

IBL can be enabled by setting the string option "environment lighting mode" to the value "light" (further modes will possibly be made available in the future). The "environment lighting quality" <float> string option with a floating value in range from 0.0 to 1.0 can be used to control the quality of the image, with 1.0 been the highest quality.

If used, mental ray would create an internal light of a "user" type and use this light in the light lists according to the conventional lightlist inheritance rules. mental ray kernel would also presample the environment, create a lookup acceleration structure and use importance driven adaptive sampling to obtain the environment light to the scene.

IBL can be combined with progressive rendering,

Feature improvements:

For finalgather precomputations, the progress report percentage messages are not joined for all passes.
Added new include file intrinsics.xmsl. For scenes using intrinsic MetaSL shader nodes as parts of shading trees and .mi files phenomena, this file need to be to generate internal intrinsic shaders declarations.
The mi_color_filter_lookup function called on an image with a negative filter value would return unfiltered value of a single image pixel. This allows to combine unfiltered image lookup in existing shaders.
For MetaSL shaders, added support for filter types "none" (unfiltered) and "data".
Shader libraries specified on mental ray command line have been loaded before parsing of the rayrc config file, changed to take config file into account.
For mi_query modes miQ_GROUP_KID, miQ_GROUP_NKIDS, miQ_GROUP_LABEL, mIQ_GROUP_DATA, and miQ_MERGE_GROUP check that the database tag passed is of the miSCENE_GROUP type and return false otherwise. This fixes potential crashes in case of inappropriate i_querymi_query usage.

Bug fixes:

Platform specific: if cached framebuffer are used, create the temporary file with FILE_FLAG_DELETE_ON_CLOSE flag to assure proper file deletion if case of application abort or crash.
For mental ray library integrations, fixed a problem in memory management. If mental ray was running low on memory and a memory block allocation failed, in some cases mental ray kernel failed to call the application callback registered with mi_mem_set_flush_failure_cb().
On Windows platforms fixed possible crash in rendering abort if memory-mapped or cached framebuffers were used (regression in version 3.7.50.5).
Debug version only: removed incorrect assert in rendering of some large scenes with BSP2 acceleration.
For MetaSL C++ backend, fixed output types for shaders with multiple outputs.
For MetaSL CIL backend, a crash in compilation of a second shader.
For MetaSL CIL backend, pass shader input parameter values properly.
For MetaSL CIL backend, fixed compilation of variable declarations of types bool, int, float and float3.
For MetaSL CIL backend, fixed compilation of Light_iterator and Sample_iterator.
For MetaSL CIL backend, fixed compilation of Trace/Occlusion/Irradiance_options and most of their methods.

Shader changes:

For production shaders, added new shader mip_rayswitch_stage. It takes one of several color parameters named after mental ray rendering stages (tile rendering, finalgather precomputing, globillum/caustic photon emission, lightmapping etc.) and returns one corresponding to the current rendering stage.
For production shaders, added new shader mip_fgshooter. This special lens shader simplifies precomputation of finalgather map for animation renderings. For tile rendering, it just passes the mi_trace_eye call with no further effects. For the finalgathering precomputation stage though, it takes the array of matrices passed as parameter and applies them as camera transformations. When, it shots an eye ray according to each of the transformations. That way, mental ray kernel registers importance of irradiance on the surfaces visible to any of the camera transformations and computes finalgather points accordingly. The "mode" parameter controls for the pre-visualization of finalgather precomputations. For the mode value 0, views from multiple cameras are painted in the same image side-by-side. For the mode value 1, an average image from multiple cameras is displayed. For the mode value 3, the image for the camera in the scene DAG is displayed (note that this mode also influences finalgather point distribution). The mode parameter is experimental and may change in the future versions.

Changes in version 3.7.50.10.

New features:

Initial implementation of the MetaSL CIL (.NET) backend on Windows platforms.
Integrated MetaSL compiler does not generate binary code representation itself: instead, the C++ backend generates C++ code and an external C++ compiler and linker is invoked to generate a DLL linked back into mental ray process. Though compilers are typically installed on Linux and MacOSX systems, this is not the case for Windows platforms. In order to avoid the requirement on the external C++ compiler with corresponding installation and license issues, the CIL backend is currently being developed.

The MetaSL CIL backend generates .NET (CIL) assembly code. An external Microsoft .NET 2.0 (or optionally a higher version) CIL assembler executable is invoked on that .cil file and generates a managed code DLL. The assembler is a part of Microsoft .NET Framework redistributable package, so no installation of a C++ compiler is required.

To enable CIL backend, the {_MI_REG_METASL_CIL} mental ray registry variable need to be set to any non-empty value. By default, the C++ MetaSL backend is used. MetaSL using CIL is supported on Windows platforms only.

Current version is extremely incomplete and can be used for demonstrational and workflow experimental purposes. Only simplest shaders like constant color shader would work in the current version. variables (l In particular, integer and float type [fixed in version 3.7.50.11], generation of state functions (like trace()), writing to state variables (like normal for bump shaders), free functions, functions with varargs, Trace/Occlusion/Irradiance_options, structs, arrays, color/vector component selections and swizzles have not been implemented yet.

Note also that the (temporary) DLL files generated by CIL backend use __stdcall linkage convention. Current version of mental ray does not expect this convention unless it knows that the DLL just has been generated by it. An attempt to load such generated file in a different mental ray process would cause a crash.

Feature improvements:

Added a possibility to control the frequency of percentage progressive messages. The -v / -verbose command line option gets an optional second argument specifying the frequency in percent. For example, -v 5 10 command line would result in printing progress messages for approximately each 10 percent. Note: this command line option has been replaced with -progress_frequency n options in version 3.7.51.0.
Added function mi_get_metasl_interface to get new mi::shader::Metasl_interface class with API for generated MetaSL shaders. This API is not intended to be used by shaders written in C/C++. The getMetaSLState method of the mi::shader::Interface class is removed. This provides a clean separation of C/C++ and MetaSL interfaces.
For the BSP2 acceleration, significantly decreased the amount of memory used for BSP tree construction on 64-bit machines.
For the BSP2 acceleration, use the approximate number of triangles per object if provided by .mi file / translator. This improves BSP2 performance vs. memory consumptions for scenes with placeholder objects and with a lot of objects consisting of few triangles.
For MetaSL shaders, the {_MI_REG_METASL_DEBUG_CPP} registry key added in version 3.7.0.8 is disabled. Instead, the {_MI_REG_METASL_TRACE} registry key is now supporting t disable the removal of the temporarily generated C++/CIL files and emit more debug information.
For mi_kdtree_lookup function, allow miKdtree_lookup_callback to be null, which may be useful if only the effective radius of the lookup is of interest.

Bug fixes:

Fixed possible pyramid texture filtering artifacts in the rasterizer. The miState::raster_x/y coordinates in the lookup were 0.5 (sub)pixel off the corresponding miState::point, which sometimes led to a wrong pyramid level selected.
For OpenEXR images, fixed possible wrong image output types, such as single channel floating point (s_fp) used instead of half (s_h), as well ass rgba_h used instead of rgb_h.
Messages from MetaSL integration are printed for "RMSL" module instead of "LIB".
For MetaSL shaders, resolved the error message on unknown depth of field ("state variable 37").
For malformed scenes containing light lists with instances referring to non-existing lights, fixed possible null tag access fatal.

Shader changes:

For multipass rendering, fixes possible artifacts in metallic paint shader.

Known limitations in this revision:

For MetaSL shaders, message passing across trace() and other functions has been temporarily disabled in order to unify C++ and CIL backends. Restored in version 3.7.51.0.

Changes in version 3.7.50.9.

Feature Improvements:

Added mi_query mode miQ_FUNC_PARAM_LOOKUP. A pointer to the string (char *) containing the name of the shader parameter, and the type of the parameter (miParam_type enum) should be passed as varargs of mi_query. The result argument pointer of mi_query is initialized with the shader parameter value if the named shader parameter is found and the value is not connected to be an output from other shader.
For MetaSL shaders sampling lights, implemented get_light_parameter() method.

Bug fixes:

For the BSP2 acceleration, use static BSP2 trees if the shutter interval is zero, even if motion vectors are required.
For the finalgather interpolation lookup to a restricted finalgather point set introduced in version 3.7.50.4, fixed the miIRRAD_DEFAULT macro to initialize the id option properly. This fixes unpredictable finalgather lookup results for shaders compiled with shader.h header file taken from versions 3.7.50.4 to 3.7.50.8.
For finalgathering used with BSP acceleration and volumetric shaders, fixed possible shading artifacts due to miState::dist not initialized to zero if finalgather ray misses geometry.
For OpenEXR output images specified with mental ray 3.5 output compatibility syntax, fixed broken compression type always set to "rle".
For the SPM-protected version of mental ray: fixed possible failure to acquire necessary number of licenses if more than one SPM server host has been specified, one of the SPM servers crashes or shuts down and none of the other SPM servers has sufficient number of licenses so that the licenses need to be collected from multiple hosts.
Fixed possible crash or allocation of large memory block fatal on loading a MetaSL shader.

Changes in version 3.7.50.8.

Feature Improvements:

For ccmesh objects with large faces, implemented pre-splitting of such faces for more efficient tessellation and displacement.
For MetaSL shaders, added {_MI_REG_METASL_DEBUG_CPP} registry. If set, the temporary generated C++ files are left over (the behavior of version 3.7.50.7 and prior), otherwise the C++ files removed. (Starting with version 3.7.50.10, use the {_MI_REG_METASL_TRACE} registry for this).
For the progressive rendering DISP interface, no longer send frame end/begin event between subsequent passes.
Added support for cached framebuffers in the progressive rendering mode. The combination is not recommended for performance reasons though.

Bug fixes:

For scenes with assemblies and incremental changes, fixed possible crash at the beginning of the second frame.
Fixed possible crash in incremental changes of scenes with assemblies.
For rendering with anisotropic BSDFs, fixed wrong orientation of highlights.
For MetaSL shaders in windows platforms, fixed embedded manifest.
For MetaSL shaders, fixed a fatal of 1D texture was used.
Platform specific: on Linux in the binary echo mode fixed broken echo of vector data.

Changes in version 3.7.50.7.

Feature Improvements:

For MetaSL shaders, the library paths (default and specified with the {_MI_REG_LIBRARY} registry and -ld_path command line option) are passed to the linker automatically. On Windows platforms it is recommended to add a path to the directory containing the shader.lib file in order to avoid specifying the full file name with the -ld_libs command line option.

Bug fixes:

Fixed a possible crash in the integrated version of raylib.
For MetaSL shader compilation, added quotes around include and link paths in the compiler invocation. This resolves MetaSL compilation errors if default installation has been applied on Windows platforms.
For MetaSL shader compilation, do not protect compiler/linker command name with the quotes. On Windows platforms, this resolved the compiler invocation failure by CMD.EXE shell, which printed "The input line is too long" error message.
For the new temporary framebuffer file handling added in version 3.7.50.5, fixed possible cleanup failure if the rendering abort was handled by the net thread.

Changes in version 3.7.50.6.

Bug fixes:

Fixed possible artifacts and NaN warnings for rendering of scenes with anisotropic BSDFs, such as Ashikhmin and architectural.
For MetaSL shaders, fixes the scheme for temporary file names. The old one allowed only a limited number.

Changes in version 3.7.50.5.

New features:

Added progressive rendering mode. There are two versions of this mode planned, for usage with mental ray conventional shaders for maximal compatibility with conventional mental ray rendering, and with build in BSDFs for maximal performance / fastest response.
This version of mental ray enables the mental ray shader compatible mode. This mode is compatible with finalgathering (in particular, with progressive finalgather precomputation), photon mapping and other techniques.

In the progressive mode, samples are placed sparsely (few samples per pixel or even one sample for several pixels), and the result is displayed immediately if an interactive display using the DISP protocol is attached. For the fastest response times, the first pass is usually computed with less than 1 sample per pixel. Additional samples are computed in the subsequent passes and the accumulated framebuffer tile is re-displayed interactively.

User may decide to stop rendering once the desired quality is reached.

Progressive mode can be enabled with the string option "progressive" on, or with the command line option -progressive on (default: off).

The "progressive max samples" <int> string option and -progressive_max_samples <int> command line option can be used to specify the number of samples per pixel after which rendering will stop automatically.

The "progressive subsampling size" <int> string option and -progressive_subsampling <int> command line option (default: 1) can be used to specify the sample density in several initial passes. If set to N, approximately one sample for each N*N pixel tile per pass will be used.

Feature improvements:

For the BSP2 acceleration, avoid loading placeholder objects during construction of the initial master BSP tree.
For the BSP2 acceleration, added dynamisation of memory consumption for large scenes. More compact (but slower for intersection test) representation of the triangle trees is enabled once there is a mental ray scene database flush for the given frame.
Added support for BSDFs in for the Irradiance Particles algorithms. In the current version, the underlying model for diffuse interaction is always lambertian. Using BSDFs give a moderate performance improvement due to the significantly reduced number of shader calls. BSDF translucency (diffuse transparency) is supported, however the classical transparency is not. The BSDFs are used automatically if attached to the material slot.
Adds new mi_query mode miQ_BUMP_BASIS, which computes bump basis vectors at the current intersection point. The miState parameter should be current intersection state, and the tag parameter should be 0. If the result out parameter is zero, the bump vectors are calculated and stored in the state->bump_x/y_list. If result out parameter is not 0, it should point to a miVector[2] array (6 floats), and the fifth argument specifying the texture space should also be provided. The result array is filled with the bump basis in addition to state->bump_x/y_list in that case.
For integrations of raylib: added two new mi_rc_run_query modes miRCQ_DISP_NET_PORT and miRCQ_DISP_TALK_PORT. The output result parameter of the function should be a pointer to int. It is filled with the port used by mental ray for the DISP net (used bu imf_disp utility) and DISP talk interactive display protocols respectively. (Note: mental ray uses the same port ids for TCPv4 and TCPv6 protocols).
The DISP talk protocol has been extended with the command "stop". This command is intended to be used with standalone mental ray application and results in mi_par_abort call for the conventional rendering, resp. in stopping rendering of further tiles and writing out of the images rendered so far in the progressive rendering mode described above.
For the imf_disp utility on Linux and Windows platforms a "Stop Rendering" menu item has been added. If selected, imf_disp sends the "stop" command for the current rendering of the image being displayed.
On Unix/Linux platforms, improved cleanup of temporary framebuffer files in case of mental ray killing/crashing: the temporary files are kept open but unlinked immediately, leading to the automatic removal by the file system.
FILE_FLAG_DELETE_ON_CLOSE flag is also added on Windows platforms, however it is less reliable.
For raylib integrations, added API functionality for MetaSL shaders (in addition to the $include instruction in .mi files):
- miMetasl* mi_api_metasl_begin(char *name) begins MetaSL shader block declaration. The name parameter is API name of the metasl code block.
- miBoolean mi_api_metasl_filename(char filename*) instructs mental ray kernel to take the MetaSL code from the specified file.
- miBoolean mi_api_metasl_code(char code*) instructs mental ray to treat the code string as MetaSL code.
- miTag mi_api_metasl_end(void) finishes the MetaSL code and adds it the the mental ray database. The tag returned is the named API tag for the block.
Known limitation: deletion of the MetaSL code block is not fully functional.
For MetaSL shader developers: added the registry {_MI_REG_METASL_TRACE}. If its value is set, mental ray will print verbose information for MetaSL shader generation useful for debugging purposes.
For MetaSL shaders, implemented one and three-dimensional texture samplers.
For MetaSL shaders, added support for the light space.
For MetaSL shaders, implemented IES/Eulumdat light profiles.
For MetaSL flattened shading graphs coming from mental ray, fixed structure member output connectivity.
Modified policy for caching memory mappable textures. Large, tiled or different endian/image type .map textures were already cached (not memory mapped). For other .map textures memory mapping used in the beginning of the rendering switched to texture caching after a specific texture was flushed and remapped 3 times. The new policy is to switch to texture caching for all .map textures after the first mental ray database flushing for the current frame. (Note: filtered .map textures with no pyramids in the file are still loaded into memory).
Improved performance of searching for files in the path list.

Bug fixes:

For the BSP2 acceleration, fixed possible artifacts or crashes due to wrong triangle indices used in scenes with multiply-instanced objects containing large number of triangles.
Put some of the memory blocks in BSP2 acceleration and Irradiance Particles under mental ray memory management / statistics.
For ccmesh objects with displacement, apply vertex displacement for the estimation of subdivision level to be used. This solves possible shading artifacts for object areas close to the camera plane.
Fixed possible crash for scenes with assemblies when incremental changes have been applied to the scene.
Removed unintended dependency of caustic photon map on importon map. This improves the visual quality of caustics.
Fixed a file descriptor leak in scenes using IES or Eulumdat light profiles. For scenes with many or with changing lightprofiles this could cause severe problems due to running out of file descriptors.
For MetaSL shaders, fixed possible artifacts due to wrong default parameter values used.
Fixed echo of default shader parameter values for parameters declared textually after an array one.
Fixed broken echo of object faces.
For the new progressive finalgather precomputation, fixed possible artifacts due to improper placement of adaptive points. The amount of jittering is slightly increased to improve the distribution quality. Fixed progressive visualization artifacts on the edges of pseudo-tiles. Fixed intensity artifacts for pseudo-tiles showing both objects with diffuse material and background environment.
For the new progressive finalgather precomputation, fixed statistic info messages.
For the new progressive finalgather precomputation, fixed possible missing DISP tiles due to a wrong number of tiles expected.
For the new progressive finalgather precomputation, fixed possible crash for rendering second and subsequent frames.
Platform specific: on Windows (32 and 64 bit), the vararg argument for mi_query mode miQ_TEXTURE_DIM has not been passed properly to the kernel, which caused crashes on 64-bit platforms. Fixed.

Shader changes:

The base shader mib_illum_ward_deriv is extended to support derivatives on all geometry types, using "bumps" (texture derivatives) if available for NURBS and polygonal surfaces and surface derivatives for subdivision surfaces and NURBS. The source code of the get_derivs() function is a recommended illustrative example for shader writers.

Changes in version 3.7.50.4.

Feature improvements:

Multi-pass finalgather precomputation visualization has been implemented.
The finalgather precomputation image report can be enabled with the -finalgather_display on command line option or with the miRENDER_DISPLAY_FG bit in the rendering mode mask. Unlike in mental ray 3.7 and prior, the computation is split into several passes, each one refining the previous ones. The finalgather precomputation respects the order of tile computations specified with mi_rc_set_taskorder() function call.

The number of FG precomputation passes can be specified with the string option "finalgather passes", or with the command line option -finalgather_passes <int>. The default value is 3.

In the current implementation, the maximal value of finalgather passes is 8. Only 2 most recent one apply adaptivity in the raster sampling, the purpose of others is to provide progressive rendering report. I.e. for the number of passes larger or equal to 3 the final image does not change.

For compatibility with mental ray 3.7 and prior, single pass finalgather progressive display can be enabled with the string option "finalgather precomp legacy".
For finalgathering, implemented a way to restrict finalgather map lookup to a specified subset of finalgather points.
In particular, this feature may be used for rendering of objects with multiple semi-transparent surfaces where explicit separating of layers is more efficient than limiting interpolation distance or increasing the density of the finalgather map.

The miIrrad_options structure passed to mi_compute_avg_radiance() function has been extended with the integer field "id". The default value for id is zero. The id is stored with each finalgather point. For interpolation, the id passed to the interpolation call is taken into account: if zero, all finalgather points are potentially accepted. If non-zero, only finalgather points with the same id are accepted.

Backwards and forwards binary compatibility of mental ray shaders and finalgather map files is maintained.
For the imf_disp tool, added -layer <name> command line option for displaying of OpenEXR files. When rendering OpenEXR images containing multiple framebuffers, mental ray stores framebuffer images in channels called <framebuffer name>.R, <framebuffer name>.G, <framebuffer name>.B and <framebuffer name>.R. In order to display one of those framebuffer images, the <framebuffer name> without suffix could be used with the -layer option.

Bug fixes:

For MetaSL shader compilation on Windows, remove manifest files form the temporary directory.

Changes in version 3.7.50.3

Feature improvements:

For multiply-instanced objects inside of assemblies, share the tessellated geometry for non-view-dependent objects. Note that sharing objects need to be a part of the same assembly.

Added new carpaint BSDF with the arguments as below:

declare shader color "builtin_bsdf_carpaint" (
    color       "base_color"       default 1 0 0 1,
    color       "reflectivity0"    default 0.1 0.1 0.1 1,
    color       "reflectivity90"   default 0.3 0.3 0.3 1,
    scalar      "fresnel_falloff"  default 5,
    scalar      "metallic"         default 0.3,
    color       "metallic_color"   default 1 1 1 1,
    scalar      "metallic_falloff" default 8,
    color       "edge_color"       default 1 0 0 1,
    scalar      "edge_falloff"     default 2,
    scalar      "unit_scale"       default 1,
    scalar      "sparkle_weight"   default 0.05,
    scalar      "sparkle_falloff"  default 1,
    integer     "mode",
    array light "lights")
end declare

For MetaSL shader compilation, the MI_DLL_IMPORT define on the compiler line is not longer required, moved to the include headers. The RAY define is no longer used and is removed.
For MetaSL shaders compiled on Windows platforms, include manifest file into the compiled DLLs. This may be necessary when the compiler version installed on the user system differs from one used to build mental ray.
For MetaSL shaders, the C++ compilation and linkage is now done on demand. For shader declaration, parsing and analysis of the MetaSL source code is done when MetaSL file is loaded.
For network rendering, compilation of the MetaSL shaders on the slave/satellites is enabled.
Work in progress: for shading trees defined in mental ray scene files, implemented the first version on the shading tree flattening, which combines shaders by connecting parameters, and compiles the composite shader. The flattening is possible for shading trees and mental ray phenomena completely consisted of MetaSL shaders (i.e. no mixture with C/C++ shaders). The flattening can be enabled by setting the {_MI_REG_METASL_FLATTEN} registry to a non-empty value.
Known limitations in this version:
- In the current version, the feature is provided to the compilation testing only. mental ray kernel does not call the flattened shaders (Fixed).
- Textures are not supported for the flattened shaders yet (Fixed).
- Flattened shaders are compiled immediately, not on demand (Fixed).

Bug fixes:

For OpenEXR headers, fixed possible crash when reading images of types a, a_h, a_fp, m, n and z.
Export mi_api_get_assembly_root function properly for geometry shaders.
For MetaSL shaders, apply include search path for the includes from MetaSL (.msl/.xmsl) files.
For MetaSL shaders: command line used for compiler and linker invocation has been limited to 512 bytes, fixed.
For MetaSL shaders: use state of the art compiler and linker options on MacOSX platforms.
For MetaSL shaders, do not echo derived shader declarations.

Changes in version 3.7.50.2

Feature improvements:

For the BSP2 acceleration, added functionality to reduce allocation of large amount of memory during the construction of the BSP2. For large scenes this may prevent mental ray from running out of memory for geometry where the kernel heuristic estimating the number of triangles in a voxel fails by an order of magnitude.
Removed dependency of the MetaSL shaders compiled to C++ on the STL.
For the MetaSL shader compiled with C++ compiler on windows platforms for mental ray debug executables: added /debug and /Z7 compiler/linker keys to keep the debug information.

Bug fixes:

For rendering of scenes with assemblies and hair, fixed possible missing hair artifacts.
During the importon emission from the camera, avoid calling miCamera::volume shader. Importons need to be proceeded with photon volume shaders, not with a volume shader attached to the camera. In particular, this fixes possible empty irradiance particle map issue.
For the MetaSL shader compilation added missing MI_DLL_BUILD define passed to the C++ compiler.
For the debug version of mental ray compiled on windows platforms, compile shaders with /MDd key. This fixes the message box pop-up when a MetaSL shader was used with mental ray debug executable. Added /debug and /Z7 compiler/linker keys to keep the debug information.
Fixed a rare crash in the object splitting code coming from out-of-range array read.
Fixed running out of memory in parsing of surfaces with very high number (50000) of trimming curves.
Fixed several memory leaks in MetaSL compiler and its integration.
Fixed a linkage errors if Access_bsdf was used in a shader.
Ignore empty strings passed to the C++ compiler (an empty string cleans up C++ compiler include path list).
For MetaSL shaders, added missing eye ray type.
Fixed a crash in mental ray when writing out images of "vta", "vts", "tag" or "bit" types into OpenEXR files.

Changes in version 3.7.50.1

New features:

MetaSL compiler has been integrated into mental ray.
If a file with .msl or .xmsl extension is included ($include statement in .mi file), it is treated as a MetaSL shader source. mental ray would read it, parse and convert it into the internal representation. This internal representation is when used to create a shader declaration and to generate shader C++ code. The external C++ compiler and linker are invoked in that file and create a dynamic linked library in the directory containing temporary files. That DLL is when linked into mental ray.

For windows platforms, it is planned to generate the managed code directly, which would remove the requirement on the C++ compiler installed.

In order for C++ compilation to succeed, the C++ compiler need to be pointed to the include directories containing mental ray headers files (shader.h and others), as well as MetaSL C++ header files (root directory for mi/metasl_runtime/*). In order to simplify it, mental ray now applies the mental ray kernel include path to the compiler invocations automatically. I.e., the include paths may be specified on the mental ray standalone command line option -I, or with {_MI_REG_INCLUDE} registry.

On windows platforms, shaders need to be linked against shader.lib or ray.lib. If shader.lib is not present in the compiler library search path, it has to be specified with -ld_libs and/or -ld_path command line options.

Known issues and limitations:
- For windows platforms, it is planned to generate managed code for MetaSL shaders directly to avoid requirement of C++ compiler installed. This is work in progress.
- It is planned to collapse shading trees consisting exclusively of MetaSL shaders at compilation time, this is not fully implemented yet. (Partial implementation exposed in version 3.7.50.3. This is not a default mode in mental ray 3.7+.)
- Network rendering with MetaSL shaders will not work yet (fixed in version 3.7.50.3)
- Multiple MetaSL features are not or partially implemented yet (fixed, with few exceptions).
- C++ and .xmsl/.msl files are not removed from the temporary file directories.
- Unnecessary compiler invocations are done to assure functionality testing. This may result in reduced performance (fixed).
- MetaSL compiler memory is not cleaned up properly at shutdown (significant memory leaks fixed).

Feature improvements:

Changed transformation inheritance for leafs inside of assemblies. Assembly leaf instance transformation is now multiplied properly with the object leaf instance transformation inside of assembly.
For the command line options, "-I" include paths are passed to the C++ compiler invoked for MetaSL and C/C++ source code shaders. If multiple -I command line options are passed to the compiler, they are concatenated into semicolon separated list, which corresponds to the intuitive behavior. mental ray 3.7 and prior replaced the old command line option with a new one.
For compilation of shaders on windows platforms, it is no longer required to add WIN_NT define to the C/C++ compiler invocation.
Changes the order of temporary directory lookups. The environment variables TMPDIR, TMP, TEMP are checked in that order. If not set, the current directory is used on Windows, and /usr/tmp is used on other platforms.
On Windows, temporarily created object and DLL files contains now the process id to avoid possible conflicts from several mental ray processes running on the same machine.
Added string option "finalgather normal tolerance". It specifies the maximal angle in degrees up to which a finalgather point normal may deviate from the surface normal for the finalgather point to be considered for interpolation.
If the string option is not present, or exceeds its valid range of 0-90 degrees (exclusive of 0 and 90), the built-in default of 25.842 degrees is used to maintain backwards compatibility.
Added a progress percentage message for optimization of Irradiance Particle maps with large number of particles.
Added compatibility version of the architectural BSDF. Its functionality is the same as for the builtin_bsdf_architectural, but the declaration follows those of the architectural material.

Enabled builtin_bsdf_architectural and builtin_bsdf_mirror BSDFs with the declarations as below:

declare shader color "builtin_bsdf_architectural" (
    color       "diffuse_refl"           default 0 0 0 1,
    scalar      "diffuse_refl_scalar"    default 1,
    scalar      "diffuse_dev",
    color       "specular_refl"          default 0 0 0 1,
    scalar      "specular_refl_scalar"   default 1,
    scalar      "refl_gloss"             default 1,
    boolean     "metal_material"         default on,
    color       "specular_trans"         default 0 0 0 1,
    scalar      "specular_trans_scalar"  default 1,
    scalar      "trans_gloss"            default 1,
    color       "diffuse_trans"          default 0 0 0 1,
    scalar      "diffuse_trans_scalar"   default 1,
    scalar      "anisotropy",
    scalar      "anisotropy_rotation",
    integer     "mode",
    array light "lights")
end declare

declare shader color "builtin_bsdf_mirror" (
    color       "reflection"             default 0 0 0 1)
end declare

The builtin_bsdf_architectural is a simplification of the architectural material, containing the parameters relevant for the material properties. The builtin_bsdf_mirror is a simple specular reflection, which multiplies the reflected color with the reflection parameter.

The mi_ambient_occlusion function behavior has been slightly modified. When the number of cache points in the miAmboccl_options argument structure passed to the function is 0, the cache is ignored and the ambient occlusion value is computed. Negative cache points value (default: -1) still can be used to take the number of cache points from the global "ambient occlusion cache points" string option.
Added support for displacement of primlist objects with no vertex normals specified.
Added A_H and S_H image types to represent alpha and intensity images in half float resolution.

Bug fixes:

For scenes with multiply-instanced assemblies, fixed motion transformations.
For scenes with multiply-instanced assemblies, fixed view-dependent approximations.
For scenes with multiply-instanced assemblies, fixed displacement mapping inherited from a material defined on an instance in the main scene and inherited into the assembly.
For scenes with assemblies, fixed mi_point/vector/normal_to/from_* function family. If called on a hit point inside of an assembly, these functions correctly use world space instead of the assembly local space now.
For builtin BSDFs, fixed missing contribution of environment lighting in glossy reflections.
Fixed artifacts in contour rendering used in combination with hull objects.
Do not echo built-in BSDF shaders.
For the BSP2 acceleration, fixed a regression from mental ray 3.6+ with memory allocations bypassing mental ray memory management. This fixes wrong memory usage statistics and possible crashes for large scenes not fitting to the memory completely.
For the BSP2 acceleration, reduced effective memory consumption by releasing memory used for tree constructions immediately after the construction is finished.
For the BSP2 acceleration used with motion blur, fixed possible missing geometry artifacts originating from numerical precision issues.
For hair objects with per-hair textures, fixed mi_query mode miQ_NUM_TEXTURES to return correct value and avoid unnecessary tessellations.
For the rasterizer used with motion blur, initialize miState::time properly for volumetric and environment shading.
Fixed a possible crash if hair was used in combinations with assemblies.
Fixed a possible crash in rendering hair with rasterizer.
Fixed a possible crash and the ray library shutdown if an extremely low memory limit was set at startup.
Fixed possible crash if Irradiance Particles were used in combination with BSP2 acceleration.
Fixed echo of verbatim textures (-echo textures command line option) for images using with more than 8 bits per pixel per color channel, such as HDR.
Fixed a crash in photon map diagnostic mode rendering used in combination with cached framebuffers.

Shader changes:

For integrators, added an externally visible mia_tonemap_api_untransform_color function which may be used to revert the tone mapping effect.
For the architectural material shader, fixed possible rare freeze in ambient occlusion computation.
For the mip_gamma_gain shader, added gamma correction for negative numbers. In particular, this protects against possible NaN values.
Fixed possible max version check warning for the subsurface scattering shader library.

Changes in Version 3.7.1.13.

Feature improvements:

For the map containers, use shorter more expressive names for the classes.

Bug fixes:

For multihosted rendering with integrated version of raylib, fixed slave host removal with the mi_msg_remove_host() function call for IPv6 addresses.

Changes in Version 3.7.1.12.

Feature improvements:

For secondary finalgather diffuse bounces, rays are traced now as finalgather rays and finalgather flags on objects an instances are respected, instead of reflection/retraction rays and flags used prior.

Bug fixes:

Fixed unnecessary large number of sub-objects for polygonal objects with displacement presampling and large maxdisplace value specified. In particular, large number of subobjects resulted in slow lightmap rendering.
For the BSP2 acceleration used with motion blur, fixed possible artifacts due to self-intersection precision issues.
For Catmull-Clark objects, min/max subdivision levels in approximations were ignored, fixed.
For Catmull-Clark objects, take into account shrinking when estimating subdivision level. This fixes unnecessary over-tessellations in such cases.
Fixed syntax error in echoing of instance override approximation lists.

Shader changes:

Make multiple-output mi_metallic_paint_shader_x work if used as an a single shader or a phenomena root.

Changes in Version 3.7.1.11.

Feature improvements:

Add new combined approximation style for shared displacement without Hermite blending. The displacement approximation styles syntax includes now "fine shared", "fine nosmoothing", and "fine shared nosmoothing". For the integrators, the miAPPROX_STYLE_FINE_NO_SMOOTHING_CONT_DISP value has been added to the miApprox_style enumeration.
The registry _MI_REG_SHARED_DISPLACEMENT which could be used to enabled the "shared" mode globally for all objects is deprecated, but could be used in mental ray 3.7 for compatibility reasons.
For the Catmull-Clark meshes, improved texture seams quality by implementing spline-rule evaluation.

Bug fixes:

For Iff image file format, enable to writing out of alpha ("a") files.
Fixed major memory leak at the end of the frame if BSP2 acceleration was used in combination with assemblies. Some of the acceleration structures have not been deleted.
Fixed a memory leak if hair rendering was enabled.
For raylib integrations: fixed a typo in mi_par_aborted() check which prevented application callback from been called.
Options consistency: disable irradiance particles if ray tracing is switched off.
Fixed possible crash in photon emission on scenes with assemblies.
For the rasterizer, environment shaders were called with non-zero miState::pri, fixed.
For ambient occlusion with enabled caching for scenes with motion blur, fixed visual artifacts due to computing of all ambient occlusion values at shutter open time.
For hair objects with view-dependent approximations, set the view dependency flags properly.
For Catmull-Clark meshes, fixed possible crash for objects with derivatives.
Platform specific: on Windows platforms, fixed IO error reporting related to the image processing.

Shader changes:

Changed the multi-output version of fast subsurface scattering shader misss_fast_shader_x to follow the same conventions as multi-output architectural material shader. The specular and diffuse properties of lights are adhered in the same way as for the architectural material shader.
For the fast subsurface scattering shader, importance-driven sampling for glossy reflections has been added, which may significantly improve performance for some scenes.
For the fast subsurface scattering shader, fixed possible artifacts edges of objects with very small scatter radii below 1 pixel in the lightmap resolution.
Added new multi-output mi_metallic_paint_x shader identical to the metallic paint shader, but allowing multiple outputs following the architectural material conventions.
Changed architectural material shader to avoid assertions in the debug version of mental ray due to of mi_ward_anisglossy function with inconsistent arguments. The change also lead to a minor performance improvement.
For the production mip_mirrorball and architectural portal light mia_portal_light shaders, changed errors to warnings in accordance to the common conventions for recoverable scene issues.
The architectural portal light mia_portal_light shader is made compatible with point-style area lights with no direction specified.

Changes in Version 3.7.1.10.

Bug fixes:

Fixed wrong motion bounding boxes for Catmull-Clark meshes with multiple motions steps and no motion vectors specified.

Changes in Version 3.7.1.9.

Changes affecting compatibility with older versions:

The shader interface is now defined in the mi::shader_v3 namespace. A namespace alias for mi::shader is added, so that no changes to the shader code are required. Shader writers are expected to continue to use the mi::shader namespace.
This fixes possible crashes (mismatched symbol linkage) if several debug shader libraries using different versions of mi::shader::Access_fb / Edit_fb were used in the same rendering in version 3.7.1.8.

Changes in Version 3.7.1.8

Feature improvements:

For the shader interface, performance of the string options lookups (mi::shader::Options::get) and accessing of the framebuffers (mi::shader::Access_fb / Edit_fb) has been improved significantly. Accessing string option still includes a hashtable lookup, so for performance reason it is recommended to access them in the init shader once.
For the Map containers, avoid rebuilding of the internal lookup acceleration structure when this is not necessary.

Bug fixes:

For scenes with assemblies, avoid unnecessary loading of assemblies with hair rendering enabled.
Fixed artifacts in the shared -displacement mode enabled with the {_MI_REG_SHARED_DISPLACEMENT} registry.
Fixed echoing of registries specifying kernel modifications including like the shared displacement trigger above.
Fixed echoing of displacement approximation for polygons and trilists.
Fixed a memory leak in the string options interface.
Fixed $ifeq preprocessor instructions of the .mi parser for the environment variables.
Platform specific: on Windows, fixed possible failure to save large photon or finalgather maps onto network drives.

Changes affecting compatibility with older versions:

The mi::shader::Interface version has been increased to 3. Shaders compiled for mental ray 3.7 alpha versions need to be recompiled. Shaders compiled for mental ray 3.6/3.6+ keep the limited binary compatibility with mental ray 3.7 over the mi::shader::Interface version 2.
For the shader interface Maps, fixed some const qualifiers.
The miObject_type enumeration is changed to keep the same values as in mental ray 3.6+ in order to include binary compatibility with mental ray 3.6+ (in particular, for the production mip_binaryproxy shader).

Shader changes:

Improved performance of the portal light shader by avoiding multiple calls for insignificant lighting.
Made production mip_binaryproxy shader compatible with mental ray 3.7.
For mia_roundcorner shader, fixed possible crash if the shader attached to the bump parameter returned struct consisting of several outputs.
For the architectural environment shader mia_envblur, improved the behavior of the shader in the erronous situation of the parameter environment not set (meaning usage of the camera/material environment shader), and the share itself being set as such shader, which has caused recursive shader calls.
For the architectural material shader, the ambient occlusion functionality of the shader has been changed to ignore objects not casting shadows.

Changes in Version 3.7.1.7

Changes affecting compatibility with older versions:

Typedef BSDF Type to miUint to avoid possible incompatibilities on sizeof(enum).
For the Map containers, adjusted signatures of get and set methods using miMatrix. This is required to as miMatrix is is typedef to float[16] and compiler cannot distinguish between reference to it and pointers for scalars on function calls.

Feature improvements:

Added a warning if deprecated version 1 file format used in mental ray version 2 and earlier is parsed.
Improved robustness of geometry tessellation if corrupted geometry data is parsed.

Bug fixes:

For multihosted rendering, fixed possible crash at shutdown if the BSP2 acceleration was used.
Platform specific: on Windows, fixed possible failure to initialize library if DNS server resolved IPv6-style address lookups (like "::1") into IPv4 addresses.

Changes in Version 3.7.1.6

Tutorial on usage of irradiance particles has been added.

New features:

Added new object flag "triangle estimate <int>" to the .mi syntax, and the new member field miObject::tri_est. If specified this field may be set by a translator or exporter and give approximate number of triangles expected for the given object. CAD application frequently would know the precise number, but even if it is know only up to an order of magnitude it is advisable to provide it. In the future versions, mental ray will use it as a hint for optimizing performance and memory usage.
For mental ray integration and geometry shaders API, added a new function
mi_api_rename(char *target, char *source)

and corresponding .mi syntax

rename "target" "source"

This function may be used to rename existing scene entities without recreating them, in particular this may be used by application interface to follow user interaction.

The function has following limitations:
- Renaming across scopes is not supported, by design.
- Compatibility with incremental echo is limited. mental ray kernel does not maintain list of renaming. If renamings are done and scene changes are echoed, application should print a set of rename commands as previously echoed parts of the scene may contain old names.
- API Notify callback is not called for the function.
- Renaming PHCR GUI is not supported.
- Renaming of shader declarations is possible bus discouraged as corresponding functions are linked using the same name as in the shader declaration.

Feature improvements:

In the displacement tessellation, the warning on removal of triangles of (close to) zero area is printed only once per sub-object and gives the number of triangles removed.

Bug fixes:

On Windows platforms, fixed broken shader.lib proxy for shaders calling to mi_scene_recreate on miSCENE_NBOX type.
For the standalone, fixed wrong process exit value 1.

Changes in Version 3.7.1.5

The manual for this version has been improved.

New features:

Syntax and application API has been added for the Map containers introduced in version 3.7.1.2.
The syntax is similar to user data elements. The layout of the map can be declared with
```
    declare map "name"
        [ dim int , ]
        paremeter1,
        ...
        paremeterN
    end declare
```
with each of parameters being one of
```
    integer   "parname"
    scalar    "parname"
    vector    "parname"
    color     "parname"
    transform "parname"
    array int integer "parname"
    array int scalar  "parname"
```
The allowed parameter types are 32-bit integers, single precision floats, float 3d vectors, 4 component float colors and 4x4 float matrices, as well as fixed size arrays of integers and floats.

For a declared layout, a map content may be specified as
```
    map "mapname" "layout"
    (
        ["mapfile1", ... "mapfileN"]             |
        { <rec1field1>, ... , <rec1fieldM> },
        ....,
        { <recKfield1>, ... , <recKfieldM> }
    )
    end map
 
```
i.e., either a list of file names containing binary map data is specified, or the list of fields for each particles in the map.

Each recSfieldT entry should consist of the proper number of space separated float or integer values. The first record in each row should be N floats corresponding to the position of a particle in the N-dimensional map.

The Map type has been added to the list of legal shader parameter types, following usual conventions. A Map cannot be an output of a shader.

The equivalent API functionality is provided with the mi_api_map_* functions family.

Additional documentations and tutorials will be added in upcoming releases.

Feature improvements:

For the BSDFs introduced in version 3.7.1.4, the BSDF component type has been transformed into a bitmask.

Bug fixes:

For the DDS image file format, non-square images with mipmap pyramids were not loaded. Fixed.
Fixed a crash if an empty photon map was loaded from a file and the finalgather fastlookup option has been enabled.

Changes in Version 3.7.1.4

New features:

Added Bidirectional Scattering Distribution Function (BSDF, or BRDF for non-transmitting materials) support for mental ray kernel.
In the current version, mental ray supports Lambert, Phong and Ashikhmin BRDFs, which are built into the kernel. Built-in shaders implementing these BSDFs are pre-registered in the mental ray API as:
```
declare shader color "builtin_bsdf_phong" (
    color       "diffuse",
    color       "glossy",
    scalar      "exponent",
    array light "lights",
    integer     "mode" )
    version     1
end declare
```
```
declare shader color "builtin_bsdf_lambert" (
    color       "diffuse",
    array light "lights",
    integer     "mode" )
    version     1
end declare
```
```
declare shader color "builtin_bsdf_ashikhmin" (
    color        "diffuse",
    color        "glossy",
    scalar       "exp_u",
    scalar       "exp_v",
    array light  "lights",
    integer      "mode" )
    version      1
end declare
```
Further standard BSDFs will follow.

An instance of a BSDF shader with given parameters may be specified with the usual syntax. Input parameters of a BSDF shader can be specified as the result of other shaders, allowing conventional shading trees and phenomena to be used. The miMaterial structure has been extended with a bsdf slot, which may specify an instance for one of the BSDF shaders listed above. The .mi syntax is extended with the bsdf keyword, allowing to specify the BSDF shader in the material.

A material shader may access the BSDF attached to the current shading state by using mi::shader::Access_bsdf class. The API for interaction with BSDFs is declared in the new shader_bsdf.h header file. It is described in the doxygen documentation in the manual.

BSDFs may be split into up to 6 components (reflection and refraction, with diffuse, glossy, and specular components each). The eval function evaluates all BSDF components, or a subset thereof, for a pair of incoming and outgoing directions. The Bsdf interface also provides a sample function, which allows importance sampling of outgoing directions for a given incoming direction. In addition, this function provides information about the sampled component and the sample weight. The latter can easily be combined with Russian roulette for absorption, which could be used e.g. for writing photon shaders.

A BSDF shader may be attached as material shader into miMaterial. In this case, mental ray takes care of simulating a material shader based on the BSDF.

With BSDF shaders, it is possible to omit photon shaders. For materials with a BSDF shader, but no photon shader specified, mental ray automatically uses the BSDF shader and use its sample function to simulate photon shader behavior. For the future version, this fallback into BSDF will be enabled for other indirect illumination techniques.

Bug fixes:

Fixed $ifeq preprocessor instructions of the .mi parser using registry variables. The mi_string_substitute function applied to a string consisting of a single reference to a undefined registry returned the unchanged string instead of empty string.

Changes in Version 3.7.1.3

Feature improvements:

For SPM-protected version of mental ray standalone executable, added a boolean string option -maxlics. This option may be usable for renderfarms running multiple mental ray executables on the same machine. If set on, independently on how many threads are running by each executable, the maximal number of licenses applicable for the given machine is allocated and shared between all mental ray executables.
Assure that miState::dot_nd initialized by intersection code, and the dot_nl parameter returned by light loop are in the [-1, 1] range. Small deviations were possible due to precision issues.
For mental ray standalone executable, changed exit code to 1 if (some of) .mi file(s) specified on the command line is not found. If multiple files are specified, rendering stops after the first file missing.
Restored scene extent info message at the beginning of rendering as known in mental ray 3.5.

Bug fixes:

For the debug build only, fixed an erroneous assert in rendering of shadowmaps.
For rendering of lightmaps, avoid allocating of unnecessary large memory blocks. For objects with high number of certifies this may reduce memory consumption significantly.
Fixed several bugs in the adaptive tessellation for Catmull-Clark meshes introduced in the previous version.
For the finalgather automatic mode, fixed possible dark artifacts on objects containing finalgather points with significantly different normal orientations (like thin objects with indirect illumination computed on both sides, or on the object corners).

Changes affecting compatibility with older versions:

The content of mi_shader_if.h header file now include several partial header files according to the features if the mi::shader::Interface class. Shader writers are still supposed to include the mi_shader_if.h file into the shader C++ files.

Shader changes:

For the architectural environment shader mia_envblur added a detection of the invalid usage of the "environment" argument pointing back to itself and for the similar case when mia_envblur shader is part of the global environment map, and the "environment" parameter is not specified, which efficiently resulted in self-reference. An error message is printed and the environment shader is disabled now.

Changes in Version 3.7.1.2

New features:

Implemented adaptive tessellation for Catmull-Clark meshes. Each region can now have a different refinement level. t-vertices with neighbor regions having different refinement levels are fixed with triangle fans. Positions are always computed on the limit surface.
Initial implementation of Map containers has been added. Maps are particle systems in N dimensional space. Each particle has a position and set data assigned.
mental ray kernel provides functionality creating, accessing and editing of Maps, as well as loading them from files and saving them to files. An efficient nearest neighbor lookup functionality is implemented, utilizing either euclidian distance or arbitrary distance functor provided by user. The distance functor may implement rejections by returning the miHUGE_SCALAR value.

Particle data may be any structure consisting of the common types including scalars, integers, vectors, colors and transforms. All particles in a single map have identical layout.

The position of a particle is an N-dimensional float vector, where N varies in range from 1 to 6 in the current implementation.

In this version, Maps need to be created or loaded by applications or shaders. API interface and .mi syntax for named Maps will be added in the next version. Minor modifications for the lookup interface are still expected. Documentation for the feature will follow. The interface functions can be found in the mi_shader_if.h file.

Feature improvements:

For OpenEXR textures, added ability to use display window. If _MI_OPENEXR_WINDOW registry is set to "display" (valid values are "display" and "data"), the OpenEXR display window used. Note that display window is not compatible with mipmap pyramids for filtered local textures.
For the integrated versions of raylib, the emergency exit for renderings running out of memory has been modified. The function mi_rc_run now returns to the application in one second and does not wait until all rendering threads are blocked. The rendering threads are still blocked on memory or database operations, possibly after mi_rc_run returns. Previous synchronous mode where mi_rc_run returned first after all threads get blocked was less robust: threads blocked on internal locks were not counted, which resulted in a possible application freeze.
For the BSP2 acceleration, more improvements in memory consumption and construction time has been applied, with focus on motion blur scenes. Several bugs have been fixed.
For string options, allow shader to access boolean values as integers. This is handy if shaders use variables of type miBoolean, which is typedefed to int.

Bug fixes:

For framebuffers, fixed possible crash if image file name with no extension was used.
For scenes with assemblies, fixed flag inheritance from the main scene to assemblies. This resolves some cases of unnecessary loading of assemblies when they are not required for rendering (for example, a combination of assemblies visible only to shadow rays and shadow option set to off).
For the '{{' and '}}' escaping for file name lookups introduced in version 3.6.52.5, fixed possible failure to apply the unescaping exactly one time.
For texture caching, fixed possible issue of not releasing references to texture pages fast enough (slower than allocating new references) for large scenes with many textures and database flushing active.
For OpenEXR images texture caching, fixed a possible crash for files with tiles not aligned the pixel coordinates origin.
For multiply-instanced trilist objects with view-dependent displacement, fixed wrong identical displacement quality used for all instances.
Fixed a bug where BSP2 acceleration could miss intersections if a substantial amount of secondary rays were spawned as children of a single eye ray. For cached glossy rays in the architectural material shader these missing intersections became visible as white dots or blocky artifacts.
Fixed wrong motion blur for BSP2 acceleration used with multiply instanced moving objects.
For integrated versions of mental ray: fixed possible crash at aborting of the rendering for large scenes which did not fit into the memory completely and triggered database flushing. The crash was more likely with BSP2 acceleration used.
In some cases when using the rasterizer with shutter_delay > 0, with motion on, and samples_motion > 1, incorrect motion could be calculated. Fixed.
In the subdivision surface compatibility mode (enabled with MI_SURF_CCMESH environment variable), fixed broken motion vectors for multi-segment motion.
For scenes with camera motion and scanline mode off, fixed missing motion blur regression from version 3.7.1.1.
Platform specific: on Linux and MacOSX, fixed possible freeze for very long rendering sessions occurring after 4 billion of database creation and editing operations.
Platform specific: on Windows, fixed a bug in rayserver service causing possible connection failures.

Changes affecting compatibility with older versions:

Changed default value for importon density used in combination with Irradiance Particles. The new value is 0.1 (per pixel). In combination with photons, the old default value of 1 is still used.

Shader changes:

For the mip_render_subset production shader, fixed possible wrong subset of geometry rendered in the finalgather precomputing mode.
Fixed rendering with mip_motionblur and mip_motoin_vector production shaders used with orthographic camera.
For the production shaders mip_mirrorball and mip_greyball, fixed possible crash with non-zero blur and no texture attached.
For production mip_binaryproxy shader, the amount of displacement has not been taken into account for bounding box computation, fixed.
For the architectural material library,transparent portal lights have been fixed to not show up in the Z buffer anymore.
The physical "transmat" shader did not support segment shadow mode. Fixed.

Changes in Version 3.7.1.1

New features:

Hair approximation. In addition to the old approximation syntax which directly specifies the number of approximating segments, it is now possible to specify adaptive length, distance, angle approximation including view-dependent approximations. Parametric approximations depending on the degree of the hair curves are supported as well.
Especially for curved hair and hair at higher viewing distance, the new approximation criterion allows to have high approximation resolution restricted to the area where needed. The lower number of hair segments result in reduced memory consumption and increase rendering speed.

The approximation can be specified on a hair object with the following syntax:

approximate hair <curve_approx>

with <curve_approx> being any curve approximation technique, for example:

approximate hair view length 3 distance 0.1 angle 5 0 7

In order to get benefit from curvature-dependent approximations, hair needs to be modelled appropriately with hair degree > 1.
For moving polygonal objects with displacement, added functionality to automatically control the quality of displacement according to the amount of visual motion.
For fast moving objects, images with comparable visual quality may be achieved with fewer displacement tessellation details compared to static or slow moving objects. Though it is also possible to tweak displacement approximation on a per-object basis, this requires a lot of manual efforts and is rarely done in practice. Per-object granularity may also be not fine enough for object having different amount of motion in different parts. The new approximation motion factor provides an automatic way for adjusting the displacement quality according to the amount of motion for a given object part.

For view-dependent fine poky displacement the adaptive subdivision checks motion length in screen space. The measured motion length is used to modify the use of the approximation constant. Geometry is reduced only in areas of the object with strong motion.

The string option"geom displace motion factor" f can be used to modify the amount of geometry reduction compared to static case, with float f >= 0. The factor f=0 turns off the feature. The factor f=1 is currently the default, and gives higher values of f give more reduction. This default value is experimental and is rather conservative: the simplification of geometry will have an effect for motion of approximately 16 pixels. In order to get effect on slower motion, higher values should be used. For example, the factor value of 8 would reduce geometry in areas of objects moving at the speed of 2 pixels per frame.
Experimental: In mental ray 3.6/3.6+ the only object-size independent approximation technique is the angle-base approximation which in some cases does not result in performance-optimal meshes. A new distance-ratio based approximation technique has been implemented, which is driven by the approximation error along an edge and the length of the edge.
In the current version, it can be enabled by setting the environment variable MI_DISTANCE_RATIO and using the distance approximation technique. Recommended values are of order 0.02, with finer approximations corresponding to smaller values.

Feature improvements:

Significant algorithmic changes have been applied to the BSP2 acceleration construction:
- for the scenes with motion blur, the construction time has been decreased dramatically
- for static scenes, memory required for the construction has been reduced and the construction time significantly decreased
- for both static and motion BSP2 trees, the quality of the constructed trees has been improved significantly, resulting in the performance improvements of the intersection tests.
Irradiance Particles have been extended with IBL-style functionality which can be enabled with the string options "irradiance particles indirect passes" is set to -1.
If Irradiance Particles interpolation is disabled then only environment presampling map is build and no further precomputation steps are required.

If Irradiance particles Interpolation is enabled, then particles are emitted in the precomputation pass in a usual way, but used as interpolation points only.

In both cases, only the environment map lighting but not diffuse bounces are taken into account.
For the detailed shadow maps saved to a file: render and store all shadowmap tiles (not just actually used ones).
The OpenEXR version as been updated to 1.6.1.

Bug fixes:

For Irradiance Particles, contribution of the environment was off by factor PI, fixed.
For volume shaders used in segmented shadow mode with autovolume turned off, added missing call of the volume shader for the last segment (one between the last occluder and the light).
For standalone version of mental ray, fixed wrong number of licensed used by machines with hyperthreaded CPUs if the number of licenses was specified as a command line parameter.

Changes affecting compatibility with older versions:

mental ray 3.7 code has been switched to use STLPORT compiled in the proprietary namespace MISTD. This resolves potential symbol conflicts for integrating application and 3rd party plugins; requirements on the compiler compatibility / version match are reduced.
OpenEXR library is linked statically. On an attempt to link mi_openexr.dll, a warning message is printed.
Removed obsolete syntax and API interface for implicit patches.
Removed obsolete box type from the miGeoBox structure.
miMaterial structure has been extended with bsdf slot.
The layout of miHair_list structure has been extended with approximation. Unused integer approximation has been removed.
Map containers have been added to the mi::shader::Interface class.

Changes in Version 3.6.52.10

Feature improvements:

Bug fixes:

For multihosted rendering, fixed a crash for scenes with assemblies.
Platform specific: on Windows, fixed a bug in rayserver service causing possible connection failures.

Shader changes:

Avoid potential linkage problems for function gloss_to_shine used in both production and architectural material shader libraries.

Changes in Version 3.6.52.8

Bug fixes:

Fixed possible crash in creating lightmap using BSP2 acceleration.

Changes in Version 3.6.52.7

Bug fixes:

Fixed a crash if irradiance particles were used with BSP2 acceleration.
Fixed wrong default value for the number of irradiance particles used fro interpolation.
For rendering of multiple frames, fixed possible unknown tag fatal in geometry tessellation.

Shader changes:

For the production shaders, clipping planes were ignored. Fixed.
For the production shaders, objects set to "invisible" blocked the view to the desired object. Now "invisible" objects are traced through properly.
For the production shaders, setting a particular object incorrectly rendered all instances of that object, not just the selected one. Fixed.

Changes in Version 3.6.52.6

Bug fixes:

Platform specific: for 32-bit Linux platforms, fixed possible artifacts due to precision issues for the BSP2 acceleration used in combination with mi_trace_continue function.

Changes in Version 3.6.52.5

Feature improvements:

For search paths, treat {{ and }} substrings as escaping for { and } symbols respectively.

Bug fixes:

Fixed a crash if cached framebuffers were used for rendering of small images with less than 4096 pixels (x_resolution times y_resolution).
Fixed mental ray standalone help banner giving wrong default memory limit on 32-bit platforms. The default value is 1000 MB on 32-bit platforms and unlimited on 64-bit platforms.

Changes in Version 3.6.52.4

Bug fixes:

Fixed echo of framebuffers for images with compression type specified.
For Irradiance Particles, fixed minor area estimation, resulting in a slightly improves image quality and performance for some scenes.
For the BSP2 acceleration, fixed possible artifacts due to bounding box intersection failures for rays almost parallel to coordinate planes.

Changes in Version 3.6.52.3

Bug fixes:

For the integrated version of raylib only: fixed possible freeze for safe exit on out-of-memory fatals (miABORT_RAY_FREEZE flag for mi_par_abort).

Changes in Version 3.6.52.2

Bug fixes:

For OpenEXR images, compression has not been enables by default since version 3.6.50.17, fixed.
For the BSP2 acceleration, fixed possible crash for motion scenes with zero-length shutter interval (i.e. shutter open == shutter, shutter open > 0).
For the BSP2 acceleration, fixed memory allocation statistics.

Changes in Version 3.6.52.1

Bug fixes:

For the RLA image format, fixed output of color + Z depth images.
For user area lights sampled by finalgather rays, do not enforce single sample per light. This may conflict with user light shader aim.

Changes in Version 3.6.50.21

Bug fixes:

For the BSP acceleration used in combination with segment or sorted shadows, fixed missing geometry normal transformation for polygonal objects without vertex normals.
For the rasterizer, fixed possible crash if mi_query miQ_NUM_TEXTURES mode was called from a volume shader.
For the BSP2 acceleration, fixed possible crash in rendering of subsequent frames for scenes with motion blur.

Shader changes:

For the architectural sky shader, fixed missing Pi factor when computing Zenith luminance from the diffuse
horizontal illuminance.

Changes in Version 3.6.50.20

Bug fixes:

Fixed missing ray-traced hair in assemblies.

Changes in Version 3.6.50.19

Feature improvements:

Added new function mi_phen_call and a set of macros
- mi_call_boolean(param)
- mi_call_integer(param)
- mi_call_scalar(param)
- mi_call_vector(param)
- mi_call_transform(param)
- mi_call_color(param)
- mi_call_tag(param)
These function and macro are equivalent to mi_eval, but do not use caching mechanism. If a shader calls this function on a parameter which is connected, then all shaders in the sub-shader chain (only) will be forced to re-compute and the result value will be returned. The original cache is not affected nor destroyed by this function, meaning, that any previous or following mi_eval() call on the same parameter will use the original cache. These function and macro are handy for evaluation shaders with temporary modifications to miState variables, such as normal or texture coordinates.
Platform specific: on 32-bit Linux platforms, compile using _FILE_OFFSET_BITS=64 define in order to allow files large than 2GB.

Shader changes:

For the architectural physical sun shader, the illumination mode was not initialized in the shader implementation, fixed.

Changes in Version 3.6.50.18

Feature improvements:

The following registries are filled in by mental ray kernel in addition to those introduced in version 3.6.50.8:
- {_MI_RAY_MAJOR_VERSION} contains mental ray (i.e. "3" for this release)
- {_MI_RAY_MINOR_VERSION} contains mental ray (i.e. "3.6" for this release).
- {_MI_RAY_BRANCH} contains the branch version (i.e. "3.6.50").
- {_MI_RAY_SYSTEM} contains a short sting for the current system class (like "windows").
- {_MI_RAY_OEM_ID}, if set, contains an ID of OEM partner.

For integrators: added a new function mi_mi_parse_assembly. This function accepts full .mi syntax and may be called for parsing of assembly files from procedural assembly callbacks only. The mi_mi_parse function accepts only a subset of the .mi syntax for the OEM raylib versions.

Bug fixes:

For the BSP2 acceleration used in combination with rasterizer or scanline, for scenes multiply-instanced objects, fixed possible self-intersection artifacts.
For hair rendering, some memory allocations were not included into memory statistics, fixed.
Fixed a tag leak for assembly filenames.
For database dumps, fixed a name for the assembly tag type.
For integrators: fixed a typo in mi_par_aborted() check which prevented application callback from been called.

Shader changes:

For the paint shader, added missing edge_factor parameter to the declaration.

Changes in Version 3.6.50.17

Bug fixes:

Fixed a tag leak if an assembly tag has been deleted with mi_api_delete function.
Fixed wrong compression options used for OpenEXR output images.

Changes in Version 3.6.50.16

Bug fixes:

Fixed a potential crash if one of the IPv4 or IPv6 interfaces for localhost was not available.
Fixed a crash if a state shader called in charge of a geometry shader accessed miCamera::buffertag.
For the BSP2 acceleration, fixed possible artifacts produced by mi_trace_continue function.
For BSP2 acceleration used in combination with assemblies, fixed a crash introduced in the previous version.
For OpenEXR files, fixed a crash if more than 8 images were stored in one file. The fix is in the mi_openexr.so library.
Platform specific: on MacOSX 64-bit PPC machines, fixed broken mi_openexr.so.

Changes in Version 3.6.50.14

Bug fixes:

For ccmesh objects fixed possible texture seams artifacts.
Fixed a crash for lightmaps computed with the BSP2 acceleration.
Fixed a crash if a combination of the BSP2 acceleration, rasterizer (or detailed shadowmaps) and hair was used.
Fixed missing shadows if a light sample loop was called from a volume shader and the BSP2 acceleration was used.
Fixed a typo in the scene options info banner.

Shader changes:

For the architectural physical sun shader, added parameters allowing to set direct normal illuminance explicitly.

Changes in Version 3.6.50.13

Bug fixes:

Fixed motion transformation inheritance for assemblies attached to a moving instance.
For the BSP2 acceleration, fixed possible failure to trigger motion oversampling.
Fixed a crash for lightmaps computed with the BSP2 acceleration.
For hair rendering, fixed possible shadow artifacts.

Changes in Version 3.6.50.12

Bug fixes:

For the BSP2 acceleration used in combination with segment or sorted ray-traced shadows, fixed possible infinite self-intersection loop leading to out-of-memory total. The underlying precision issue has been observed on 32-bit Linux platform only.

Some of memory allocations in the BSP2 acceleration were not counted in mental ray memory statistics, fixed.

For the BSP2 acceleration used in combination with motion blur and ray traced shadows, fixed possible failure to detect motion and trigger motion blur oversampling.

Changes in Version 3.6.50.11

Feature improvements:

New caching mechanism for building of the BSP2 tree with motion is enabled. The BSP2 construction time for large scenes is significantly reduced.
Added IPv6 support for mental ray multihosting rendering. In the .rayhosts file, as well as on the command line, IPv6 addresses may be specified. Host names and IPv4 addresses are still supported. IPv6 addresses need to be put into the ][ ] brackets following common URL convention. Note that if slave host is specified on the command line, the [ ] brackets may need encapsulation depending on the shell used.
mental ray master host tries to establish connections to slaves using either TCP or TCPv6 protocol, whatever is available and applicable to a slave host (if a slave host is given by name, both TCP and TCPv6 protocols are tried out, the order depends on operating system used and OS configuration).

Bug fixes:

Fixed null tag fatal in direct creation (not output of rendering framebuffers) of OpenEXR image files, used in particular for lightmap textures.
For multi-imaged OpenEXR output files, removed unnecessary conversion to color for depth, normal and motion framebuffers which caused loss of some information.
Fixed a crash if photon emission with photon autovolume mode enabled has been used in combination with BSP2 acceleration.
For the BSP2 acceleration, fixed possible crash if mixed static and moving geometry was used.
For raylib integrations, fixed wrong thread counting in out-of-memory emergency exit.
An application may register a memory callback which is called when mental ray kernel is not able to free sufficient amount of memory itself. If this callback is called and the application is aware that it is not able to free a significant amount of memory, it may decide to call mi_par_abort() function with miABORT_RAY_FREEZE flag. In that case, mental ray kernel would hang all rendering threads and delete currently not accessed database elements, freeing a moderate amount of memory. After a short delay, the mi_rc_run call will return. Application is supposed to save user work and exit. mental ray library functionality should no longer be used, in particular the mi_raylib_exit should not be called.
For multihosted rendering, fixed possible crash if multiply instanced trilist objects are used.
For Irradiance Particles, fixed possible crash in evaluation if importance was close to zero.
For Irradiance Particles, fixed computation of surface translucency.

Shader changes:

For the architectural material, fixed possible dot artifacts in rendering of indoor scenes with bright environment and glossy interpolation used in combination with "detail distance".

Changes in Version 3.6.50.10

Bug fixes:

Assure that for shadow shaders miState::scanline is 0 as specified.
For the scene echo, do not print instance light lists if echoing of lights or light instances is disabled.

Changes in Version 3.6.50.9

Feature improvements:

Added IPv6 support for the socket layer of mental ray DISP protocol (immediate display of tiles as they rendered). mental ray would listen for connections done with either TCP or TCPv6 on the same port id. The imf_disp tool would try to get the connection with any of TCPv6 and TCP protocols.
Extended progress strings enum with Irradiance Particles entry.
Added Irradiance Particles entry to the mi_query mode miQ_STAGE.

Bug fixes:

For the BSP2 acceleration used in combination with multi-segment motion, fixed a severe bug in computation of internal boxes during the BSP2 construction. This fixes possible artifacts and significantly improves performance for such scenes.
For ccmesh objects, fixed possible crash for mixed tri-/quad-meshes.
For texture caching, fixed a crash in image conversion when a color image was opened as a scalar texture.
For the command line option -ip, fixed the default value to off if the option is given.

Changes in Version 3.6.50.8

New features:

Added novel global illumination technique, Irradiance Particles (short: IP), which is in many cases superior to finalgather and/or photon mapping in terms of image quality, usability and performance.
A short description of the technique may be given as follows: before rendering, importons are shot to the scene from the camera. Their hit positions with information on the amount of direct (and possibly indirect) illumination coming at their position (hence the name "irradiance particles") are combined into a map. Optionally, one or more passes of indirect illumination can be computed. By the nature of the algorithm, the computation is importance-driven. During rendering, Irradiance Particles are used to estimate the irradiance for every shading point: if just direct illumination was collected for irradiance particles, this is equivalent to one bounce of indirect lighting. Irradiance can also be interpolated from precomputed values at particles' positions.

Irradiance Particles cannot be used in combination with globillum and finalgathering (the options are automatically adjusted on an attempt to do so). However, Irradiance Particles are compatible with caustic photons.

The following string options can be used to control irradiance particles.
- "irradiance particles" <boolean>: this string option enables the Irradiance Particles. The default is off.
- "irradiance particles rays" <int>: this string option controls the number of rays shot while estimating the irradiance. The number is similar to the number of rays used for finalgathering, but specifies the maximal number and delivers better quality compared to finalgathering used with the same number of rays. The minimum is 2. The default value is 256.
- "irradiance particles indirect passes" <int>: the number of possible passes of indirect lighting. If this number is greater than 0, then a sequence of passes is computed to collect the irradiance coming from multiple indirect illumination bounces. Irradiance Particles would have both direct illumination and indirect illumination information in that case. If this number is 0 then Irradiance Particles will have only direct illumination information. The default value is 0.
- "irradiance particles scale" <float> | <color>: this is the global scale factor applied to the intensity of the irradiance during rendering. Values different from default do not lead to a physically correct rendering but are useful for artistic purposes. A single value or a color may be specified. A single value is expanded to a color having the same R, G and B components. The default value is 1.0.
- "irradiance particles interpolate" <int> | <string>: this string option is used to control the use of interpolation. It can either be a numeric value or a string. 0 means no interpolation, 1 means interpolate always, 2 means interpolate only for secondary rays (that is, no interpolation for eye rays and interpolation for reflections, retractions, etc). Alternatively, strings "never", "always" or "secondary" may be used. The default value is "always".
- "irradiance particles interppoints" <int>: the number of Irradiance Particles used for the lookup interpolation. The default value is 64.
- "irradiance particles env" <boolean>: this flag enables the use of the environment maps for irradiance computation. In that case a separate particle map is built for the environment (if an environment shader is present) and used during rendering for image based lighting. The default value is on.
- "irradiance particles env scale" <float> | <color>: this is a global scale factor applied to the irradiance contribution of the environment. Like "irradiance particle scale" string option, a single scalar or a color may be specified. The scaling factor is relative because it applies to the environment irradiance only: the environment irradiance could be further scaled (multiplicatively) if the user specifies a global scaling factor with the "irradiance particle scale" string option. The default value 1.0.
- "irradiance particles env rays" <int>: the number of rays used for the computation of irradiance coming from the environment map. This number can be greater than the number of rays used for normal irradiance computation, especially if most of the environment is covered by scene geometry (typical case: is a room with just one or two windows). For outdoor scenes, it should work fine with a smaller number of rays. Default value is equal to the number of rays specified with the "irradiance particles rays" string option.
- "irradiance particles file" <file_name_string>: specifies the file name for the Irradiance Particle map. If the file exists, mental ray will try to read the Irradiance Particle map from it (unless"irradiance particles rebuild"string option is set on) . If the Irradiance Particle map is not (successfully) read from the file, mental ray compute it and save it to a file with the given name. This is consistent with the option for photon maps. The default value is no name, meaning that Irradiance Particle is not associated with a file.
- "irradiance particles rebuild" <boolean>: if set on, mental ray will compute the Irradiance Particle map even if a file with a name given with the previous string option exists. If it if off, then mental ray will either read it from file (if the "irradiance particles file" string option is used) or it will reuse the Irradiance Particle map coming from the previous frame rendered. The latter can be useful in animations, which are flicker-free. However, the Irradiance Particle map may loose quality if the objects and camera are moving, so this can actually be done only for fly-throughs. In addition, since the particle map is view dependent, inaccuracies might show up on the frame image borders. The default value is on.
The following command line options are equivalents of the strings options listed above: -ip, -ip_rays, -ip_indirect_passes, -ip_scale, -ip_interpolate, -ip_interppoints, -ip_env, -ip_env_rays, -ip_env_scale, -ip_file, -ip_rebuild.

A draft mini-tutorial for this feature is included in this release.
The following registries are filled in by mental ray kernel:
- {_MI_RAY_VERSION} contains mental ray version
- {_MI_RAY_PLATFORM} contains the platform name
- {_MI_RAY_DATE} contains mental ray date
- optional {_MI_RAY_REVISION}, if set, contains the internal repository version from which executable has been build.
These registries may be used to setup multiple mental ray versions and platforms (like 32-bit vs. 64-bit shaders) on the same machine.

Feature improvements:

Added a new string option "importon emitted". If specifies the total number of importons to emit and overrides the "importon density" string option. The options are similar, but "importon emitted" specifies the value which is independent of the image resolution and thus is better for tuning.
For the new framebuffer interface, added a treatment of compatibility-style file types with sign, like "+rgba". This is not intended and recommended usage, but it resolves a common mistake.
For applications integrating raylib: added a new type for progress message in ambient occlusion computation pass.

Shader changes:

Extended architectural mia_physicalsky shader with the explicit Perez model.
For architectural shaders, added mia_ciesky shader to the collection.
For the architectural physical sun shader, fixed "automatic_photon_energy" mode.
For the architectural material shader, improved performance by ignoring bumps for finalgather rays.
For the architectural material shader, fixed the code which removed specular highlights for finalgather rays.
For the architectural material shader, fixed wrong reflections of some volumetric effects.
For the architectural material shader, fixed possible artifacts from the "detail distance" feature if no environment shader was used.
For the architectural output shader, fixed a minor issue in triggering "once per render" error messages.

Changes in Version 3.6.50.7

Feature improvements:

For ambient occlusion cache, decreased reaction time on the aborting of rendering.

Bug fixes:

Fixed possible crash if finalgathering was called from a volume shader and BSP acceleration was used.
For the BSP2 acceleration, fixed possible artifacts due to a missing motion (time contrast) oversampling.
Fixed motion transformation inheritance into assemblies instances.
Fixed severe artifacts if multiply-instanced assemblies were used in combination with BSP2 acceleration.
For OpenEXR output files containing multiple images, the -file_dir command line option has not been used properly, fixed.
For OpenEXR output files containing multiple images, fixed possible crash if image type conversion for some framebuffer images is required.
Removed false error message on editing objects during rendering. The operation is legal for assemblies.

Changes in Version 3.6.50.6

New features:

Framebuffer image caching. This option may save a significant amount of memory for rendering of high resolution images. If enabled, only a small fraction of each resulted image is present in memory: newly rendered tiles and tiles recently accessed by the display callbacks, output shaders, writing out final images into files and kernel in some diagnostic modes.
Framebuffer image caching is superior to the virtual framebuffers (-fb_virtual on command line option) as for virtual framebuffers an output image could be memory mapped or unmapped as a whole, i.e. the advantage was only present if multiple framebuffers were rendered.

Framebuffer cache may be enabled with the command line option -fb_virtual cache or by setting the miOptions::fb_virtual value to 2 in the integration version. The type of the miOptions::fb_virtual has been changed to integer. Legal values are: 0 for in-memory framebuffers, 1 for memory-mapped (virtual) framebuffers, 2 for cached framebuffers.

Note that output shaders operation on usage framebuffers should use mi_img_get* and mi_img_put* functions and avoid miIMG_ACCESS and miIMG_LINEACCESS macro, as well as mi_img_access and mi_img_cache_get functions. Output shaders should also not access image->height and image->width directly, but use mi_img_get_width and mi_img_get_height functions instead.

Feature improvements:

Construction time for the BSP2 acceleration used in combination with motion blur has been decreased dramatically. In the current version, some addition memory is allocated temporarily for the construction period. The amount of that memory will be reduced in upcoming releases.

Bug fixes:

For ccmesh objects, fixed texturing on seams.
For hair rendering with the rasterizer, fixed possible artifacts due to improper miState::normal_geom initialization for hair primitives.
Fixed a string options tag leak.
Fixed a memory leak and possible crash in the Framebuffer::remove method.
In finalgather precomputing statistics, do no print out the number of light shader calls.

Shader changes:

For the base lightmap shader (mib_lightmap_write), added new fg_mode parameter controlling precomputing of finalgather points on the lightmaps. Possible values for the parameter are: -1 for no precomputation (compatibility behavior), 0 for every mesh vertex, N>0 for a single finalgather point per N pixels. The default value is 3.

Changes in Version 3.6.50.5

Feature improvements:

Performance of the BSP2 in combination with motion blur has been improved.
Do not print an API warning on a shader redeclaration if the redeclaration is done in the namespace scope, such as an "anonymous" namespace for assembly or geometry shader. This may be used for including of shader declarations to assembly files.
Added remove() method to the Framebuffer class to allow removal of a framebuffer.
For the mi_geoshader_echo_tag function, echo link statements if the nolinkcommand echo option has not been set.

Bug fixes:

Fixed a parser error for lightprofiles attached directly to miLight (i.e. not used as a light shader argument).
Fixed a parser error for lightprofiles attached directly to miLight (i.e. not used as a light shader argument).For the Framebuffer interface, get_index no longer modifies the index value to -1 if a buffer with the given name is not found.

Changes in Version 3.6.50.4

Feature improvements:

Improved overall BSP2 acceleration performance.
For the BSP2 acceleration, improved parallelization if placeholder objects are used.
Use more intuitive names in the BSP2 progress messages.
For the Framebuffer interface class, added methods get_buffercount and get_buffername returning the number of framebuffers and the name of a framebuffer with a specific index.
For the mi::shader::Options::get(const char *, int*) interface function, recognize boolean option value and convert it to 0 or 1. This fixes a nonintuitive boolean option lookup failure if a variable of type miBoolean was used.

Bug fixes:

For the BSP2 acceleration used in combination with the rasterizer and motion samples option equal to 1, build static(no motion) acceleration. This dramatically reduces memory consumption and improves performance for that option combination.
Fixed missing echo of the BSP2 acceleration.
For scenes with assemblies, fixed possible wrong motion and missing scene parts due to an erroneous assembly motion bounding box.
For scenes with assemblies, fixed a possible crash if the main part of the scene contains no geometry or geometry with no texture spaces only.
For trilist objects with displacement, fixed echoing of the displace approximation.
Fixed possible crash if the rasterizer with a lens shader was used in combination with the BSP diagnostic mode.

Shader changes:

For the architectural material shader, significantly improved performance if the segment shadow mode was used.

Changes in Version 3.6.50.3

Feature improvements:

For framebuffers, added a boolean user flag, the name is more intuitive. Its semantic is the opposite to the autofill flag: if set, the content of the framebuffer is not filled by mental ray kernel. The autofill flag still can be used as an antonym, but is deprecated.

Bug fixes:

Fixed possible crash when using displacement shaders in assemblies.
For OpenEXR files containing multiple framebuffer outputs, fixed wrong output for user framebuffers.
Fixed possible framebuffer counting problem resulted in "failed to create frame buffer" error message.
Platform specific: fixed possible parse errors for non-tagged primlist objects on platforms using GCC4 compiler.

Changes in Version 3.6.50.2

Feature improvements:

For OEM integrations of mental ray library, file assemblies are enabled.
For raylib integrations: added two new functions mi_disp_fbmap_name and mi_disp_fbunmap_name. These functions may be used by DISP callbacks and are functionally equivalent to mi_disp_fbmap and mi_disp_fbunmap, but take a name of the framebuffer instead of the index as argument.

Changes in Version 3.6.50.1

Feature improvements:

For the BSP2 acceleration, significantly improved performance and reduced memory consumptions for scenes with motion.

For the framebuffers, added a new autofill flag. If set, the content of the framebuffer is initialized by the mental ray kernel according to the framebuffer type (i.e. color, normal, z depth etc.). The default value for the autofill flag is true. The autofill flag supercedes the useprimary flag effective for the color framebuffers only. The useprimary flag is still supported for compatibility, but integrators are encouraged to use the new autofill flag instead.

Bug fixes:

For the BSP2 acceleration, fixed possible missing geometry for objects with multiple instances in the scene. Fixed possible self-shadowing artifacts for such scenes.
For the debug build only: fixed possible crash if multi-segment motion blur was used.
Fixed possible artifacts if a single eye rays spawned a very large number of rays in the shading subtree.
Fixed a crash for mi_query mode miQ_PRI_INDEX used on a miState with state->pri == 0.
For ccmesh objects with displacement, added clipping of negative displacement values smaller than -maxdisplace. Added a warning for NaN displacement values returned by displacement shaders.
For non-ccmesh subdivision surface objects, fixed missing displacement in the immediate tessellation called by a geometry shader.
For the ccmesh objects, fixed possible crash for large scenes not fitting into physical memory.
For ccmesh objects, fixed instance and instance override approximations which were not taken into account.For ccmesh objects, fixed possible wrong subdivision level selected for the immediate tessellation called by a geometry shader.
Fixed possible view-dependent tessellation quality artifacts for NURBs if orthographic camera was used.
Platform specific: for MacOSX platforms, fixed possible tessellation artifacts in the edge merging.
For the framebuffer C compatibility interface, fixed
- unintentional coping of the primary framebuffer color to user framebuffers
- broken output of user framebuffers with index 8 and larger
- echoing of user framebuffers

For the mental ray standalone, fixed non-functional -fb_dir command line option.
For the OpenEXR image files, removed a limit on the string length of the qualified file name.
For OpenEXR output images, do not write the frame buffer as a name channel for the Z depth buffer, use channel name "Z" instead.
For the pass file rendering used in combination with non-primary ("user") framebuffers, fixed an error in writing out the first rendered tile of the image.

Shader changes:

For the architectural material shader, unrolled the change in version 3.6.0.25: if the mental ray kernel trace depth limit is reached, the black color is returned instead of the environment color.
For architectural material shadow shader, significantly improved performance in case of opaque materials.

Changes in Version 3.6.3.1

The manual has been updated.

Bug fixes:

For mental ray standalone, fixed wrong command line help for several command line options.

Shader changes:

For the architectural round corner shader used in combination with bump mapping, fixed a crash if the round corner shader was called as a shadow shader. For scenes using such shader combinations, performance has been significantly improved.
For the architectural material shader, fixed possible "glowing object" artifacts in scenes using finalgathering with low finalgather trace depth limit and HDR environment shaders.

Changes in Version 3.6.2.8

Feature improvements:

For ccmesh objects with n-sided polygons (not only triangles and quads), the new fast subdivision surface implementation introduced in version 3.6.0.20 has been enabled. In addition to increased performance, the quality of texture mapping for such objects has been improved.
Added new mi_query mode miQ_PRI_ASSEMBLY_INST_TAG. For a valid intersection state, the assembly leaf instance for the hit primitive is returned. If the geometry does not belong to an assembly, the null tag is returned. This query mode may be used by shaders to implement custom attribute inheritance from the main scene file.

Bug fixes:

For multihosted rendering, fixed a rare crash observed for scenes using large bsp acceleration.
For multihosted rendering, fixed possible database statistics and dump reporting tags in the NULL module.
For rasterizer used in combination with volume shaders, fixed a possible crash for shaders setting state->pri to 0, such as the physical parti_volume shader.
For OpenEXR images, fixed a crash for single channel images containing only the "A" channel.

Changes in Version 3.6.2.7

Bug fixes:

Fixed possible artifacts for sphere and cylinder area lights originating from a wrong normal transformation.
For the BSP2 acceleration, miState::dist was not set to zero if no geometry was hit, fixed.
Fixed void photon emission from object lights.

Shader changes:

For the subsurface scattering shader, enabled the support for instance light lists.
For the physical light shader, fixed wrong energy scaling for visible rectangle and cylinder lights. This fix affects the color of ray intersections with lights only, the illumination of scenes is unchanged.

Changes in Version 3.6.2.6

Feature improvements:

For the BSP acceleration, removed unused BSP memory control from the interface. If parsed from .mi file, a warning is printed.

Bug fixes:

Fixed possible crash if the miQ_LIGHT_NORMAL mode of mi_query has been used on a cylinder light.
Fixed a crash if a combination of a contour rendering and a shader modifying miState::raster_x/raster_y was used. The contours put outside of the currently rendered tile are ignored.
For texture caching, removed an erroneous info message that conversion on the fly from one image type to an identical one is required.

Shader changes:

For architectural material shaders, disable reflection and refraction interpolation if contour rendering is enabled.

Changes in Version 3.6.2.5

Feature improvements:

For OpenEXR output files containing multiple framebuffers, changed the channel naming scheme. The channels names used for color frame buffer buffername are buffername.R, buffername.G, buffername.B, buffername.A. For the primary framebuffer, the channel names R, G, B and A are still used.

Bug fixes:

For BSP2 acceleration used to render scenes with assemblies, fixed unnecessary loading of assemblies. In addition to lower memory consumption, this fix results in significant speedup for scenes with a large number of assemblies due to a better parallelization of assembly loading.
Fixed possible artifacts for ccmesh objects: if multiple regions are created, wrong materials from other regions on the same surface may be used.
Fixed possible crash for the ray-trace shadows if the simple shadow on mode was used in combination with the BSP acceleration, and the shadow shader called ray or shadows tracing functions.

Shader changes:

Some of the contour shaders were not exported on Windows platforms, fixed.
For the architectural material photon shader, fixed possible tracing of photons in an uninitialized direction.
Improved embedded strings for shader library date/version.

Changes in Version 3.6.2.3

Feature improvements:

For scene echoing, do not add object space requirement for irrelevant declaration types.

Bug fixes:

Fixed a regression in 3.6.1.3: scenes with the combination scanline on, trace off, volume on, autovolume on options were rendered black.
Fixed possible crash for large scenes using BSP2 acceleration in a combination with a large number of assemblies.
Fixed minor memory leaks in assembly loading.

Changes in Version 3.6.2.2

Feature improvements:

For the rasterizer, improved the statistics output.
For the framebuffers, added a new useprimary keyword for backward compatibility with the C-style interface for color + Z files.
For the sake of consistency, the names for useopacity introduced in 3.6.2.1, have been changed slightly.

Bug fixes:

For multihosted rendering, fixed a crash in animation rendering. For some tags, the old content was re-transmitted over the network for the second and subsequent frames.
For rendering of scenes not fitting into physical memory using a combination of assemblies and BSP2 acceleration, fixed possible unknown tag access fatal.
For the rasterizer, fixed a performance drop originating from increased shading rate due to unnecessary increased tessellation for objects with slow motion.
Fixed a regression in 3.6.0.30: for detailed shadowmaps, fixed missing shadows if light shadowmap detail samples has not been specified.
For multihosted rendering, fixed possible freed memory access after a frame end.
For framebuffers, added missing default file format.
Fixed Color + Z combined output for Iff and RLA files using C-style buffer interface.

Shader changes:

For fast and physical subsurface scattering shaders, added instance light list support if the mode parameter is set to 4.
For the architectural portal light shader, fixed a small darkening artifact at the midpoint of the edge of the portal light if it was very close to a surface.
For the architectural round corner shader, improved the quality of edge rounding.
For the architectural material, fixed possible artifacts in Fresnel reflection (e.g. glass) in combination with bump mapping and rounded corners.

Changes in Version 3.6.2.1

Feature improvements:

For the rasterizer, add transparency/opacity handling to "userbuffers" (i.e. non-primary color buffers). In the new framebuffer API, the useopacity boolean modifier can be set on a framebuffer. For the integrated version, the "useopacity" attribute can be used.
It is also possible to use the boolean string command "rast useopacity" or the command line option -rast_useopacity to force the behavior for all color framebuffers. This way the new feature may be used in combination with the C-style framebuffer API.
For the framebuffer interface, deduce the filetype from image file extension if it has not been specified.
For free form surfaces, the performance of derivative evaluation has been improved.
Added some non-null checks to the mi_shader_if.h headers. The change is relevant only for products built without the rendering component, not for mental ray and raylib.

Bug fixes:

Fixed a possible crash in multihosting rendering when a slave was removed from the list of the rendering hosts shortly after the rendering finished.
Fixed possible texture seam artifacts in ccmesh tessellation.
For some combinations of output statements and non-primary (user) framebuffers, detect the bit-per-pixel value properly.

Shader changes:

For base shaders, added a mib_fast_occlusion shader based on the build-in ambient occlusion.

Changes in Version 3.6.0.32

Feature improvements:

For finalgathering, significantly reduced the requirement for continuous memory block sizes. This may give memory usage benefits on 32-bit machines rendering with large finalgather maps.

Bug fixes:

For TIFF image file format, fixed file header for RGB (no alpha) images.

Changes in Version 3.6.0.31

Bug fixes:

For the rasterizer, take per-object shading samples into account.
Fixed possible crash on an assembly unload.

Changes in Version 3.6.0.30

Bug fixes:

Fixed possible crash in multihosted rendering if rasterizer or shadowmaps were used.
Fixed a direct allocation of memory (a significant amount if lightmaps were used), which bypassed internal memory counts.
Avoid unnecessary memory consumption during lightmap rendering. The mi_state_set_pri function called on a lightmap mesh primitive would invalidate all previous miState::pri hit points.
Fixed null tag access fatal for ccmeshes tessellated by geometry shaders.
For orthographic cameras, fixed a wrong decision on the number of split objects to use.
Include trilists in the tessellation statistics printed at the end of rendering.

Shader changes:

For architectural material shader, fixed possible artifacts if an "ambient light" (light with a shader which sets dot_nl to 1) was used.

Changes in Version 3.6.0.29

Bug fixes:

Fixed poor tessellation quality in polygonal and free form surface object splitting for orthographic cameras.
Reduced trimming artifacts by improving numerical stability for clipping of subobject trimming curves.
Added missing echo of derivatives for free form surfaces.
For ccmeshes, added missing object view dependency flag. Fixes missing re-tessellation for rendering of multiple frames.
Fixed possible artifacts if finalgather was used in combination with the rasterizer, the BSP acceleration was used and samples motion was set to 1.

Changes in Version 3.6.0.28

Feature improvements:

For subdivision surfaces, fixed a crash by intercepting objects with bad topology (non-2-manifolds) and ignoring them.
For raylib integration: compile JPG module with C++ in a namespace to avoid potential name conflicts.

Changes in Version 3.6.0.27

Feature improvements:

Significantly improved rasterizer performance for high depth complexity scenes (such as scenes with hair) rendered with the transparency depth limit added in the version 3.6.0.23.
For MetaSL shaders, added support for shadow shaders.

Bug fixes:

Platform specific: fixed freezes in multihosting rendering with Windows slave/satellite hosts.
For scenes with a large number of assemblies, speedup assembly preprocessing.

Fixed possible hole artifacts for free form surfaces with trimming curves.

Changes in Version 3.6.0.26

Bug fixes:

Fixed a crash for probe rays and ambient occlusion used with hairs.
Added missing echo of subdivision surfaces creases and corners.

Shader changes:

For the architectural material shader, fixed a crash if multi-output material was used as a part of a phenomenon or a shading tree not on its root.
Improved documentation architectural and production shaders.

Changes in Version 3.6.0.25

Feature improvements:

Added a new mi_query mode miQ_STAGE. Shaders may use the resulting queried value to detect the current rendering stage. The shader.h file lists possible values: normal tile rendering, finalgather precomputing, ambient occlusion cache creation, photon and importon emission and lightmapping.
Decreased verbosity of the BSP2 acceleration.

Shader changes:

For the architectural material shader, fixed artifacts and possible crash for for the ambient occlusion with color bleeding mode.
For the architectural material shader, reflection and refraction return the environment color if the mental ray kernel trace depth limit is reached. The former behavior (black color) was inconsistent with other mental ray shaders. The behavior if the shader-specific trace depth limit is reached is unchanged.

Changes in Version 3.6.0.24

Feature improvements:

Partially reverted changes in the area light sampling pattern introduced in 3.6.0.22 for compatibility reasons: the new pattern must be explicitly enabled by a light shader returning value 3 (instead of miTRUE) for the first sample of the sample light loop. This way a better quality of a pattern with known number of samples is combined with a possibility to use variable number of samples in smart light shaders.

Bug fixes:

Fixed a significant tag leak which could lead to running out of tags in a long rendering sequence.
Ambient occlusion was disabled for finalgather rays in the finalgather precomputing stage. Fixed.
Fixed instance light linkage for volume shaders. The light lists are taken according to the miState::instance, not to the miState::pri which is null for the volume shaders. In order to get a lightlist on an instance, the instance object must be hit by a ray in the current rendering tile.
Fixed possible out-of-memory crash for free form surfaces objects using fine parametric approximation without displacement.
Fixed echoing of scalar textures.
Fixed a crash in rendering of merge groups.
For ccmesh, fixed a possible crash for vertices with valence larger than 64.
Fixed artifacts for object area lights.
For raylib integration only: fixed framebuffer integration issue if options, camera and rendering calls were accessed from multiple threads.

Shader changes:

For the production matte shader, fixed missing environment reflections on objects seen in reflections. Improved default shader parameter values.
For the production mip_motion_vector shader, renamed the raw_floating_point argument to floating_point_format and changed its type to integer. Valid values are 0 (no floating point is written), 1 (raw floating point pixel values are written, same as the old "on" value) and 2 (the x component takes the pixel aspect ratio into account).
For the architectural material, improved NaN reporting: only the initial one but not dependent ones is printed.

Changes in Version 3.6.0.23

Starting with this version, changes in the shader packages are given in a separate list.

Feature improvements:

For the rasterizer, added a string option "rast transparency depth" and the corresponded -rast_transparency_depth to limit the depth complexity. For high depth scenes this option may be used to significantly improve performance without noticeable changes to the rendered image.
For the rasterizer, renamed the string option "motion factor" and command line option -motion_factor to "rast motion factor" and "-rast_motion_factor" for consistency.
Reduced possible differences in the displaced geometry if the same scene is rendered twice: with and without rasterizer.
For free-form surfaces, improved performance of C1-discontinuity detection.

Bug fixes:

Fixed inheritance of shading samples defined on instances.
For subdivision surfaces, fixed an uninitialized motion bounding box which lead to missing instances.
If primary framebuffer was assumed implicitly, the user framebuffers were not created. Fixed.

Shader changes:

For the architectural portal shader, adjusted the semantics of the "multiplier" argument. In addition to the emitted light, it is applied to the visible light surface if the "visible" shader argument is set.
Changed architectural round corner shader to return vector. Added "bump_mode" and "bump_vector" parameters. The "bump_mode" defines the semantic of vector arguments and the return value. If the mode 0 (default), the shader behaves in a backwards compatible way and perturbs the normal itself. In mode 1, it accepts the "bump_vector" as argument and returns the modified bump vector.
For architectural environment blur shader, improved the quality of a strong blur.

Changes in Version 3.6.0.22

Feature improvements:

Added sharp creases to the new fast ccmesh implementation.
Changes sampling pattern for area light sources. Material shaders using mi_sample_light loop may now break the loop without introducing a major bias after an exact power of 2 samples have been taken.
For ccmesh, reduce memory consumption and improve performance for objects which are not split into subobjects.

Bug fixes:

For ccmesh, fixed possible wrong texturing at texture seams.
For ccmesh, fixed possibly wrong bounding boxes.
For the rasterizer box filter always used size 1. Fixed.

Changes in Version 3.6.0.21

Bug fixes:

Fixed a freeze in incremental changes of chained userdata blocks.
Fixed wrong maximal subdivision level clipping for subdivision surfaces parametric approximation.
Fixed a freeze at raylib exit for echo of tessellated subdivision surfaces.
For the architectural sky shader, fixed an artifact if the sun was partially below the horizon.

Changes in Version 3.6.0.20

Feature improvements:

Added a completely new subdivision surface implementation based on a mixed triangle-quad subdivision scheme. The new implementation is several times faster than the previous one.
The following approximation types are supported by the new implementation:
- parametric
- [fine] [view] length [min max]
In version 3.6.0.20 the new code path is used for triangle-quad ccmesh input objects not containing features listed below. The new fast implementation support for the these features will be added in upcoming versions:
- sharp creases (added in version 3.6.0.22)
- semisharp creases
- vertex features
- n-sided polygons (not only triangles and quads) in ccmesh (added in version 3.6.2.8).
For objects containing containing one of the above features or other types of approximations, the mental ray 3.5-based subdivision surface implementation is used.
For the BSP2 acceleration, improve performance for scenes with many small objects by improving the control of the master BSP tree depth.
Switched production shaders to use the new framebuffer interface.

Bug fixes:

For assembly placeholder callbacks, miState::global_lockwas not initialized. This completes the fix in version 3.6.0.18.

Changes in Version 3.6.0.19

Feature improvements:

Framebuffers are echoed with the new interface.

Bug fixes:

For mi_api_framebuffer_add function, allow filetype and filename arguments to be null. This way framebuffers with no image files associated to them can be declared in the C-style API.
Fixed tile artifacts for rendering of hair in combination with instance lightlists.
For hair rendering, fixed artifacts on semi-transparent hair strains shaded twice due to hair subdivision. Improved hair rendering performance.
Fixed bad syntax for user data block echo.
Internal stack overflow check was nonfunctional except for mi_call_shader function calls. Fixed.
For scenes containing both assembly and object instances, fixed missing geometry issues.
For architectural material shader, fixed wrong bump mapping for reflections introduced recently.

Changes in Version 3.6.0.18

Bug fixes:

For cameras with output shaders, fixed an unknown tag fatal in the second frame introduced in the version 3.6.0.17.
Fixed echo for camera outputs and framebuffers. The echo uses mental ray 3.5 compatibility syntax, switch to the new framebuffer syntax will follow.
For geometry shaders and assemblies call miState::global_lockwas not initialized. Fixed.
For architectural portal light shader, fixed possible artifacts if scanline rendering was used.
For architectural round corners shader, fixed possible artifacts for scaled objects.
For the architectural material shader used in the compatibility multiple-output mode, fixed a wrong connection of "specular" to "translucency".

Changes in Version 3.6.0.17

Feature improvements:

For the BSP2 acceleration, improved heuristic for the local object BSP trees.

Bug fixes:

Fixed a crash if mi_trace_probe hits an object with no material.
Framebuffers prepend miCamera::output list used for rendering and cleaned up properly after the rendering finishes. For rendering of multiple frames this fixes a possible unknown tag assess fatal.
Fixed a regression in version 3.6.0.16: user framebuffers were not created if output to a file was not specified.
Fixed a regression in version 3.6.0.16: for incremental cameras, the output statements no longer accumulate.
Fixed a crash in -finalgather_display mode.
For the subsurface shader: fixed uninitialized alpha channel for multiple-output version of the shader.
For the architectural material: fixed possible crashes and artifacts for hair and scenes containing objects without a material.

Changes in Version 3.6.0.16

Feature improvements:

Significantly reduced memory consumption for the trilist objects used with displacement or in the rasterizer.
Added a new function
```
    miBoolean mi_api_framebuffer_add(
        miTag   buffertag,
        char   *datatypes,
        char   *filetype,
        float  *fileparams,
        char   *filename);
```
which should be used for the C-style framebuffer API. The functions mi_api_output_file_parameter and mi_api_output_file_def are deprecated. The parser has been changed to use the new function.
Added an extended version of the architectural material shader, mia_material_x, which supports multiple outputs. added a bump input argument to improve integration with other shader packages.
For production shaders, added a new shader mip_gamma_gain, a simple shader to multiply and/or apply gamma to a color. Added a new mip_rayswitch_environment shader to perform choice based on the hit type (the background, or something else). Added shaders for mirror and grey ball environment maps. For matte shadow, added support for instance light lists, multiple outputs, ambient occlusion, grayscale shadow catching and so called illuminators (a second light list that actually illuminates the matte objects).
Added multiple output for the fast subsurface scattering shader.
For spectral rendering, allow to specify spectra shader arguments as an (interleaved) list of wavelength and values directly.
If an undefined registry or environment variable has been used in the string lookups (like search path lists), the complete string has been evaluated to empty string. If a registry name started with underscore, this has been done silently (debug message only). The behavior is changed to evaluate just the registry itself to an empty string. In particular, this improves usability if the .rayrc configuration file is missing or does not contain the standard registries.
For the BSP2 acceleration, added a limit of 1 Mio rays per eye ray shot. This decreases severity of a rare issue when a shader traces multiple continue and transparency rays which lead to a cyclic hits sequence.

Bug fixes:

Fixed crashes and missing framebuffers when mental ray 3.5 compatibility framebuffer API was used for user framebuffer declarations.
Fixed possible crash in rasterizer if shader set state->pri to 0.
Fixed wrong treatment of instance exclusive shadow light lists containing all lights in the scene.
Fixed possible artifacts (missed hits) for mi_trace_continue function used with the BSP2 acceleration. The offset from the initial ray origin has been applied twice.
Fixed several memory leaks.
Fixed multi-core counting for 32-bit executables running on 64bit Linux.
Fixed possible crash in rasterizer if autovolume was used.
For finalgathering, state->pri has not always been properly set to 0 if nothing was hit. This caused artifacts for some environment shaders.
Fixed a crash if finalgather was used in combination with assemblies.
Fixed possible crash if finalgather was used with lightmapping and BSP acceleration.
If ray tracing is enabled by phenomena requirements or techniques like finalgather, the acceleration is no longer set to BSP2 automatically.
For BSP2 acceleration, fixed a significant inefficiency for the BSP construction in the scenes with a large number of objects with few triangles.
Fixed possible artifacts in the hair rendering for hair shading with strong dependency on precision of the hit.
For spectral rendering, sort the wavelengths for incoming spectra.
For spectral rendering, spectra shader arguments are echoed properly now. The echoed list contains wavelengths and values as used for internal computations.
Platform specific: on Windows, fixed possible crash if mi_trace_continue is used for scenes with assemblies.
Platform specific: for GCC4 build, fixed infinite recursions and/or empty tiles in the networking rendering due to broken deserialization.

Changes in Version 3.6.0.14

Feature improvements:

For mi_trace_eye function, initialize state->pri to null before the intersection test. This fixes possible artifacts in chained lens shaders which shoot multiple eye rays. The physical_lens_dof shader behavior is not affected as the shader already set the state->pri to null before tracing eye rays.
For the BSP2 acceleration, added a hard limit on the total number of rays which could be spawned by a single eye ray to prevent deep nested recursion. Previously, only the recursion depth was controlled.
For the case when a local filtered texture does not contain pyramids and the image format has texture caching support (like OpenEXR, Iff or BMP): downgrade the message that such textures cannot be cached from warning to info.

Bug fixes:

Fixed a crash for some combinations of user framebuffer and output shaders created with the old mental ray 3.5 compatibility interface.
Fixed a crash for finalgathering used in combination with assemblies.
Some memory allocations for C++ class new operators and STL container allocators were done directly with malloc calls. Switched them to use mi_mem_allocate. This fixed some inaccuracy in memory statistics and a possible crash in malloc if the mental ray memory limit is not set.
Platform specific: fixed a crash for reading DDS images on MacOSX.
Fixed a memory leak in network transfers.
Fixed a minor memory leak in rasterizer for empty tiles.
Fixed possible crash if large bsp acceleration was used.

Changes in Version 3.6.0.12

Feature improvements:

For the integration of the library: the internal representation of framebuffer sets has been cleaned up. The tag/pointer cast on the return of mi_rc_run_query in miRCQ_FB mode should no longer be required.

Bug fixes:

Fixed possible memory leak for small trilist objects with displacement.

Changes in Version 3.6.0.11

Feature improvements:

Changed parser to accept the grammar produced by the Reality Designer™ plugin for usage in Reality Server®. Most of the extensions are ignored, the options attributes are mapped to the string options.
Improved performance of polygonal displacement in combination with the rasterizer.

Bug fixes:

Restored unintentionally removed -fb_virtual command line option.
Fixed lack of motion blur in shadows created with shadowmaps.
For incremental changes of named chained lens, volume and environment shaders in the camera, and all types of of named chained shaders in material: fixed .mi parser to avoid breaking the chains after parsing the incremental change. Note that the usage of named chained shaders is strongly discouraged as such shaders cannot be used in more than one chain.

Changes in Version 3.6.0.10

Feature improvements:

For state shaders called in miSHADERSTATE_STATE_INIT mode, miState::type is initialized correctly. This may be used by a state shader to detect photon emission, lightmap or output shader stage.
The miState::cache is no longer used for intersection optimization. The member is kept for compatibility reason, but should be considered deprecated. The marker functionality (setting miState::cache to 0 in order to allow certain shader types to call ray-tracing functions as documented in the manual) is still available. This change gives a small performance improvement.
For debug version only: improved instrumentation of the database related code.
Improved performance of mi_shaderstate_get function. This function is used heavily by some shader packages.

Bug fixes:

For multihosted rendering, fixed possible network connection failures introduced in the previous version..
For multihosted rendering, fixed network transport for the Framebuffer database type.
Fixed mi_opengl.so loading failure.

Changes in Version 3.6.0.9

Feature improvements:

Improved misss_physical shader behavior if it is called as a shadow shader. In that case, the material parameter is evaluated. This improves usability under the assumption that the shader attached to the material parameter can be used as a shadow shader.
Improved performance of the PHEN module responsible for the shader and phenomena calls.
Slightly improved performance for free-form surfaces subdivision.

Bug fixes:

Fixed a crash if photon emission for a scene with hair was used.
Fixed a stack overflow crash if a shader calls light sampling from a shadow shader (which is illegal) and segment shadow mode is used.
The mi_mem_set_allocator did not overwrite the allocation, reallocation and freeing functions properly. Fixed.

Changes in Version 3.6.0.8

New features:

Caching for the ambient occlusion has been added. The cache is created prior to the image tile rendering is several passes, each refining the previous one. The following string options cab be used for controlling the ambient occlusion cache:
- "ambient occlusion cache" enables the caching. If set to off, the ambient occlusion passes are skipped and the occlusion value is computed for each mi_ambient_occlusion function call. The default is off.
- "ambient occlusion cache density" specifies the maximal density of ambient occlusion points after all refinement passes, in points per pixel. The default is 1 ambient occlusion point per pixel.
- "ambient occlusion cache points" specifies the number of ambient occlusion points used for interpolation. The default value is 64.
The corresponded command line options are added, abbreviated to -ao_cache, -ao_cache_density and -ao_cache_points resp.

Feature improvements:

For the BSP2 acceleration, support for the motion oversampling has been added.
Performance and support for motion blur scenes with the BSP2 acceleration is improved.
Due to the nature of the BSP2 acceleration, the BSP depth and leaf size are configured automatically. The BSP diagnostic modes are thus irrelevant. They are removed for the BSP2 in order to improve performance.
For ambient occlusion "ambient occlusion samples" string option has been renamed to "ambient occlusion rays" in order to distinguish between the number of rays used to compute an ambient occlusion value, and the number of points used in the cache interpolation. The experimental string options "ambient occlusion spread", "ambient occlusion spreadexp", "ambient occlusion near" and "ambient occlusion far" have been removed. The miAmbient_occl structure, the samples member has been renamed to rays, and cache_points member has been added.
Removed quadratic complexity on the number of rendering tiles in the DISP module. This gives a significant speedup for scenes with high resolution and small tile size.
For the rasterizer and scanline rendering, significantly improved performance for scenes with displacement, there a very conservative value for the maximal displacement value was given and the displace presample option was not used.
Command line options for importons are added.
For the architectural material shader, improved the shadows quality for transparent objects. The old version would take "total internal reflection" into account, which could cause the shadow of a glass object to be completely black. The new version of the shader renders slightly lighter shadows which gives a better visual impression and is closer to the physically correct shadows. Note that the physical correct lighting could only be achieved with caustics.
For the architectural material shader, added two new modes to the thin_walled parameter that force the shader ignore the normal vectors orientation for primary rays. If thin_walled is set to 2, the object hit is treated as solid (thick walled) glass, but the surface hit by an eye ray is considered to be outside of the glass, regardless of the direction of the normal. Mode 3 is similar, the the primary ray interactions are rendered as being inside of a media, which could be useful for rendering of underwater shots.
For the architectural material shader, added support for the instance light lists, following the mode convention used by other shaders if the mode parameter is set to 4.

Bug fixes:

Added missed treatment of instance shadow lightlists for hair objects.
For volume shaders mi_instance_lightlist always returned empty light list. Fixed.
Fixed broken image output for CT image file format for the case then the frame buffer bit per pixel value differs from those requested for the output file.
Fixed a crash if grid acceleration was specified: the mapping to the BSP2 acceleration is done properly.
Fixed a crash in photon emission for scenes with volume photons.
Fixed a crash if a framebuffer without filename was specified for a camera.
Fixed wrong index returned by mi::shader::Framebuffer::get_index().
The options banner printed in the beginning of rendering lists all frame buffers, not only those used for writing to files.

Changes in Version 3.6.0.7

Feature improvements:

Multiple changes improving performance.
Changed a way color user framebuffers are clipped. Instead of using clipping identical to the main color buffer, the bit depth of each user framebuffer is taken into account individually.

Changes in Version 3.6.0.6

New features:

Added interface for named framebuffers. The camera definition in the .mi file the following statement could be added:
```
    framebuffer "name"
        type string
        filename string
        filetype string
        filtering boolean
        colorprofile string
        compression string
        quality string
        dod boolean
        dpi int
```
The new syntax replaces file output statements. The old syntax is still supported. The shader interface is extended with the Framebuffer class. In order to get the framebuffer index used by mi_fb_put and mi_fb_get functions, the Framebuffer::get_index method could be used. Framebuffer class allows accessing and editing of the framebuffer properties. In order to access or edit the framebuffer set, the smart pointers Access_fb and Edit_fb have been added.

Feature improvements:

BSP2 acceleration performance has been improved.
For the TARGA image format, set the image descriptor byte in the header of RGBA image files to '8-bit attribute data' as suggested by the TARGA specs (TARGA32 subformat).

Bug fixes:

For network rendering, fixed possible freezes and crashes if BSP2 acceleration is used.
Fixed possible shadowing artifacts if ray traced shadows were used.
Object area lights have been fixed.

Changes in Version 3.6.0.5

Feature improvements:

Added binary format for hair objects using the following syntax:

    hair
        ...
        scalar [ <number of scalars> ]
        binary `...<binary block of scalars>...`
        hair   [ <number of offsets> ]
        binary `...<binary block of hair offsets>...`
        ...
    end hair

Protect the architectural material shader mia_material and round corners shader mia_roundcorners against being called as a displacement shader.

Bug fixes:

Fixed several memory leaks.
Fixed wrong pincounting which could lead to a crash when rendering large scenes.
In rasterizer and for BSP2 acceleration, fixed possible wrong transformation of objects inside of assemblies.
Fixed possible self-intersection artifacts for the BSP2 acceleration used in assemblies.
Fixed possible wrong transformation of visible area lights.
Fixed crash in rasterizer if a shader has used mi_trace_continue function.
Fixed possible memory corruption in rasterizer filtering.
Fixed missed output if -o command line options was used and output shader statement was present.
Fixed a possible crash in volume rendering in autovolume mode.
Fixed possible shading artifacts if an orthographic camera was used with rasterizer.

Changes in Version 3.6.0.3

New features:

Assembly support. An assembly is a subscene attached to a bounding box, which is loaded on demand when the bounding box is hit by a ray, or is contained and not completely occluded in a rendering tile, if the rasterizer is used. In cases of insufficient memory, assemblies are unloaded automatically and reloaded the next time they are required.
An assembly may contain all scene DAG elements, such as objects, instances and groups, as well as textures, materials, shaders, phenomena, lightprofiles and other scene elements. Lights and cameras can be contained in an assembly, but they will be ignored in the rendering process. Thus, shaders and instance light lists should refer to the lights from the main scene only. This requirement is due to the fact that the rendering kernel would not be aware of them until all of the assemblies are loaded in order to apply lights on an object from a different assembly or from the main scene. In mental ray 3.6 lights inside assembly would not lit the geometry inside of that assembly, but this restriction may be removed in the later versions.

An assembly can be created procedurally as a result of geometry shader call, or loaded from a .mi file.

Geometry shader can register a callback which is called on demand to create the assembly context, possibly multiple time if the assembly is unloaded. The API is similar to placeholder object callback. The details are given in the Assembly API section of the manual.

In order to add a file assembly to the main scene, the following syntax must be used:
```
    assembly "assembly_name"
        box [min_x  min_y  min_z  max_x  max_y  max_z]
        [ motion box [min_x  min_y  min_z  max_x  max_y  max_z] ]
        file "file_name"
    end assembly
```
The individual parameters are:
- The box statement specifies a bounding box for the assembly, in the space of the assembly. It is the union of all bounding boxes of all geometric objects defined in the assembly transformed to the assembly space. The box parameter must always be specified since it is not possible to compute the bounding box without loading the assembly.
- The motion box is similar to the box, but contains the minimum and maximum components of all motion vectors. For assemblies with motion the motion box must be specified. Unlike object motion bounding boxes, the assembly motion bounding box should take into account not only object motion vectors, but also motion transforms inside of the assembly.
- The file statement specifies the name of the file where the subscene is stored. A list of directories for looking up the file can be specified with the -assembly_path path_list command line option, with the {_MI_REG_ASSEMBLY} registry, or with the MI_ASSEMBLY_PATH environment variable.
Assemblies files should contain a root statement at the end:
```
    root "group_name"
```
The group "group_name" would be attached to the main scene of the assembly.
Nested assemblies are not supported.

For more details see the the Assembly section of the manual.

Known limitations in the current version:
- The standard BSP tree does not support assemblies. The new BSP2 acceleration described below should be used for ray tracing.
- Unloading of assemblies is unstable. In particular, on unloads, there could be minor memory and tag leaks on anonymous shaders and materials.
- The assembly exit callback is not called.
Note: unlike objects, assemblies do not have visible, trace, shadow and similar flags enabled by default. During the scene preprocessing the inheritance rules apply up to the assembly attachment level. In particular, this means that an instance of an assembly (or some instance of a higher hierarchy level) must have the corresponded flags set explicitly. The inheritance inside of assembly is applied only when the assembly is loaded.
MetaSL support. MetaSL is a powerful shading language targeting current and future shading requirements. mental mill®, a product of mental images, can be used to create MetaSL shaders and graphically compose shader graphs and phenomena. The proprietary mental mill compiler technology generates abstract syntax tree representations of shaders and Phenomena. These are then translated to the software and hardware shaders libraries by mental mill. Software shaders are linked as usual dynamic libraries.
Note that mental mill is not a part of mental ray. The shader libraries created by mental mill are used by mental ray in the same way as is done for the "native" shaders compiled from conventional programming languages like C++.

mental ray shading functionality has been extended to allow efficient support for the MetaSL shaders.

It is possible to use a mixture of native and MetaSL shaders in the same scene. It is also possible to combine "native" software mental ray shaders and MetaSL shaders in a shader tree or in a mental ray Phenomena at rendering time.
New BSP acceleration technique. A new BSP acceleration technique has been developed for efficient rendering of large BSP trees. It may be enabled by the "-acceleration bsp2" command line option, or with the "acceleration bsp2" option in the .mi file.
Unlike the traditional BSP acceleration (which is still available), the new techniques does not require manual tuning for BSP traversal performance and memory consumption improvements. Depending on the scene, the new BSP2 tree acceleration can be faster or slower (for small scenes) than the traditional one. However, the memory requirements are much lower. Thus, the new BSP2 is recommended for rendering of large scenes, especially scenes which do not fit to the physical memory of a machine, and for scenes containing assemblies. If the new BSP2 acceleration technique is used for scenes with assemblies, an independent tree is created for each assembly. This allows the complete removal of all structures associated with the BSP if an assembly is unloaded from memory.

Limitation in the current version: for scenes with motion, time contrast oversampling is not supported yet.
Importons. Importons are "particles" similar to photons. However, they are shot from the camera and traverse the scene in the opposite order. Instead of "energy", they hold a quality which is an importance of the contribution to the final image. The information from importons is used by the kernel for distribution of rendering efforts according to the final contribution to the rendered image, i.e. so-called importance-driven sampling. Importons are thus a foundation for new view-dependent indirect illumination techniques. As a first of them, mental ray 3.6 adds an importance-driven photon map described below. Further techniques are in development.
Importon emission is done before photon emission and tile rendering. The string options controlling the importon emission are:
- Boolean "importon" enables the importon emission. The default value is off.
- Float "importon density" controls the number of importons shot from the camera per pixel. The minimal value for this option in the current implementation is 0.02, i.e approximately 1 importon per 50 pixels. The default and recommended value is 1. Lower values will speed up importon emission, but will also lead to a worse final image quality.
- Float "importon merge" is the world-space distance used to merge importons closer than that distance. The default value is 0.0, which means than merging is disabled.
- Integer "importon task size" controls the pixel size of the importon emission jobs. The default value is 64. In most cases importon emission takes much less time than rendering of the image, so this is a fine-tuning option only.
- Integer "importon trace depth" controls the diffusion of importons in the scene. If set to zero, importons will not scatter on the diffuse bounces. In some cases it may be required to use more than a single diffuse bounce. This is the case if the combination with finalgathering is used, or when the "importon traverse" option is set off.
- Boolean "importon traverse", if set on, enables a special behavior of importons traced from the camera. Such importons would not be blocked by even completely opaque geometry. Instead, they would be stored for all intersections with geometry on the ray from the camera to infinity. This leads to a significantly higher number of importons stored in the scene. However, it removes the discontinuity in the distribution of the importons originated from the visibility to the camera function.
Most of the existing photon shaders work out of the box with importons.

Limitations in the current version: importon emission does not work properly with lens shaders significantly modifying the direction of eye rays, such as a fish eye lens.
Importance-driven photon maps. A new technique based on the importon map described above has been implemented. This technique is superior to the photon merging introduced in mental ray 3.5. If importons are enabled, they are used to control the density of the globillum photon maps and to improve the lookup time.
If importons are enabled, the "globillum merge" option has no effect. The merging density is controlled by the importon merging distance instead.

Like for photon map merging, the number of photons emitted to the scene should typically be much higher than the number of photons stored in the map. Unlike for simple photon merging, there is a limit on the side of the photon map after merging: a photon is kept in the photon map only if an importon close to it could be found. This puts a natural limit on the size of the photon maps as the number of photons stored in the scene can not significantly exceed the number of importons, and the latter has simple dependency on the image resolution and "importon density" string option value.

As a bonus effect, reduced photon counts result in faster photon map lookups. Distribution of photons according to the contribution to the final image significantly improves the resulting visual quality. In order to avoid artifacts on the contours of the objects and at the edges of occlusion from the camera, it is recommended to set the "importon traverse" option on.

In the current implementation, importons are only used for the creation of photon maps. Thus, importons are discarded after the photon emission is completed.
Ambient occlusion. A native implementation of ambient occlusion has been added. A shader may call the new function:
```
    miScalar    mi_ambient_occlusion (
        miState                   *state,
        struct miAmboccl_options  *amboccl_options,
        miVector                  *bent_normal);
```
The optional bent_normal argument, if not null, will be filled with an average of un-occluded directions. The ambient occlusion options structure is the following:
```
    struct miAmboccl_options {
        int         size;         /* size of the structure             */
        int         rays;         /* number of probe rays              */
        miVector    axis;         /* the axis for the cone             */
        miScalar    spread;       /* cosine of the spread angle        */
        miScalar    spread_exp;   /* the exponent of the cosine distr  */
        miScalar    mindist;      /* near clipping distance            */
        miScalar    maxdist;      /* far clipping distance             */
        int         cache_points; /* # ao points for lookup            */
    };
```
The size argument contains the size of the structure for the current version of mental ray, allowing future extensions without breaking binary compatibility. The samples argument is the number of probe rays shoot for computing ambient occlusion at a shading point. The axis and spread arguments contain the main direction and cosine of the angle, defining the cone in which ambient occlusion samples are taken. Note that the spread value of 1 corresponds to the angle opening of zero degrees, and the value of 1 to 90 degrees, i.e. to a hemisphere. The cosine_exp argument defines the exponent used for the weighting of samples. The mindist and maxdist arguments define near and far clipping distances for computing occlusion. For efficiency, it is recommended to use reasonably small maxdist values.
A macro miAMBOCCL_DEFAULT(ao_options, state) can be used to fill the ambient occlusion options with default values.

The following string options can be used to specify global defaults for ambient occlusion options:
- "ambient occlusion rays" sets the number of probe rays shot at a shading point. The default value is 256.
Limitations in the current version: ambient occlusion caching is available starting with the version 3.6.0.8. Once added, the ambient occlusion options structure and the list of corresponded global default options will be extended. The spread value must be non-negative. The spread_exp value used for the computations is always 1, meaning ambient occlusion according to the Lambertian BRDF.
Added support for the DDS image file format for single textures and cubemaps. Internally cubemaps are stored as a single texture containing all 6 faces.
A new mi_query mode miQ_FUNC_IS_ROOT is added. The miBoolean argument is set to miTRUE if the shader called called directly by the kernel as a "root" of the shader graph or phenomenon. A shader returning a large structure could use it to prevent a possible crash from attaching it directly to a material or light.
The paint shader library with shaders for the simulation of car paint was made available with mental ray release 3.5.3.1. For documentation, please refer to the included Paint Shaders documentation.

Feature improvements:

Light lists, when created, may specify the sampling cone by providing an axis of the sampling cone and a cosine of the spread angle, like in the code sample below:

    mi::shader::Interface* iface;
    mi_query(miQ_RAY_INTERFACE, state, 0, (void*)&iface);
    auto_ptr<MI::RAY::LightList> llist(iface->createLightList(
                                          state, state->normal, 0.5));
    for (mi::shader::LightIterator light = llist->begin();
         light != llist->end(); ++light) {
        while (light->sample()) {
            miColor c = light->get_contribution();
            // ...
        }
        if (light->get_number_of_samples()) {
            // ...
        }
    }

This could be used for sampling lights from both sides of the surface without the state->pri=0 trick.

For the rasterizer, added a new string option "motion factor", float value. If set, it adjusts the effective shading rate on moving objects. It can be used to improve rendering performance for motion scenes without a visual quality decrease.
For the rasterizer, added new shading_samples field to miObject and miInstance, which allows overriding the global shading samples specified in the options. The corresponded syntax in .mi file: shading samples float.
Added the new importance field to miState. It is propagated to child rays. For rendering quality/performance improvements, a shader written for scenes using indirect illumination may multiply this field by a certain factor before calling mi_compute_irradiance or mi_trace_... function. The factor may be chosen as an intensity of the color with which the result of the function call would be added to the result of the shader, or, in case of tracing multiple (glossy) rays, as the inverse of the number of rays shot.
Added the new boolean field photons_only to the miLight structure. If set, the illumination from the light is treated as indirect. Light sampling of the light would return black. To compensate, photons emitted from the light would stored on the first bounce in addition to the second and further bounces. For a physically correct scene, the illumination model is thus the same.
This option can be used for significant performance improvement for scenes with a high number of low intensity lights: the length of the light lists, and thus the number of light shader calls and shadow rays is reduced significantly at the price of a larger photon map.

The corresponding .mi syntax in the light section: photons only boolean.
Added new finalgather mode. If the string options "finalgather mode" is set to "force", finalgather map creation and interpolation will be disabled; a set of finalgather ray will be shot each time mi_compute_irradiance or mi_compute_avg_radiance is called. The rendering speed would be much slower than in other finalgather modes, delivering a superior image quality.
For OpenEXR output files, allow to combine arbitrary frame buffers in the single output statement. Before this update, only the combination of the main color buffer and Z buffer was possible. Added the possibility to write any subset of the first 8 user framebuffers (i.e. "fb0" to "fb7") to such files. If an OpenEXR output file contains a user framebuffer and at least one other buffer, the user framebuffer channels are called "Rn", "Gn", "Bn", "An", where n if the index of the user framebuffer for color type outputs, "Zn" for Z-depth output etc.
Restored mi_img_type_query function in the shader interface. For a given image type this convenience function returns the number of component, and the number of bits per component. It also fills the optional array of 4 booleans with flags for the components if present.
For raylib integration: changed mi_disp_fbmap function to return 0 if called with out of range index. In particular this should simplify integration of the display callback in the finalgather precomputing preview mode.
Significantly improved parallel performance on some platforms with large number of CPUs and non-uniform memory access (NUMA).
In physical library, dgs_material, dielectric_material and path_material shaders get the "mode" argument, which is the same as for base shaders. The allowed values for mode are:
- Mode 0 is the compatible behavior, light lists are taken as is.
- Mode 1 applies mi_inclusive_lights
- Mode 2 applies mi_exclusive_lights
- Mode 4 forces shader to use the instance light list.
The versions of the shaders have been incremented.
Added new boolean string option "contrast all buffers". If set, the contrast oversampling test is done based on all color buffers, including user framebuffers.
Added mi::shader::Options::remove method for string options removal.
For mental ray library integrations, if the mi_api_variable_set function is called with null name and value arguments, the whole API variable table is cleaned up.

Changes affecting backwards compatibility with mental ray 3.5:

The miState->pri_idx, the index of triangle in the sub-object box, is removed. With the changes done for assembly support, the miState->pri completely identifies a primitive in a scene. Lightmap and some contour shaders should use the new miQ_PRI_INDEX query mode to obtain the triangle index. Lightmap shaders should use the new mi_state_set_pri function to set the current triangle index.
Some missing const qualifiers were added to the interface functions. Compiler warnings or errors would indicated unintentional usage and should be taken seriously. There is at least one exception though: If GCC 4 is used, it may print out warning on the mi_tri_vectors, which should be ignored.
The mi::shader:Interface class now has the new version 2. The new class is incompatible with the version 1 used in mental ray 3.5. However, the interface version 1 is still supported and can be obtained with mi_ray_interface_version, so mental ray 3.5 shader will not need a recompilation due to this change.
The camera space (in particular used in the the old .mi1 files) is no longer supported. The geometry will be loaded correctly, but the texture spaces and other coordinate systems may be wrong.
Grid acceleration is no longer supported. The main advantage of grid was the low memory footprint in heavily multiply-instanced scenes. The BSP2 acceleration technique with assemblies is a superior replacement.
In the miOptions, the fields rapid_shading_samples, rapid_collect_rate and rapid_motion_resample are renamed to more intuitive rast_shading_samples, rast_collect_rate and rast_motion_resample respectively.
For the integrators on Windows: setting binary mode on the stdin is no longer applied. If raylib is integrated into a standalone-style product which could be used in the pipe mode, the mode must be set in the integrated code explicitly with the call: _setmode( _fileno(stdin), _O_BINARY).
mental ray 3.6 keeps file outputs and output statements separately. All output image files are written out after all output shaders are applied. Writing out a frame buffer to one file, applying an output shader to that framebuffer and writing the modified buffer to a different file is not possible.

Bug fixes:

Fixed a crash in the parsing of phenomena containing named structs with names longer than one character.
Fixed a crash in combined floating point color and z-depth output ("+rgba_fp,+z" and "+rgb_fp,+z" output statement) supported for iff file format.
Fixed a memory corruption when finalgather is used in simple or multiframe mode, and mi_compute_avg_radiance is called with varying number of FG points.
In multihosting rendering, fixed a possible crash when a local texture is deleted between frames.
Fixed a crash in photon emission from rectangle and disk area lights when the projection map rejected all emission directions.
Fixed a crash in the mi_scene_growedit function called for the miSCENE_NBOX type.
Fixed a crash during tessellation of inaccurate objects where some but not all vertices have bump vectors.
Fixed a possible memory corruption in the polygonal object splitting.
Fixed a crash when the mi_sample function was used in detail shadow maps rendering
Fixed a possible crash when using the "automatic" final gathering mode and setting the number of final gathering points to 0.
Fixed possible memory corruption when reading extended photon maps for the finalgather fastlookup mode
The value of the register _MI_REG_MEMORY_LIMIT is now in megabytes.
When using rasterizer, the Z-buffer contained the depth of the first completely opaque hit instead of the first not fully transparent one. Fixed.
In the rasterizer, fixed front/back face check for object instances flipped in one axis (i.e. those with negative transform determinant).
Fixed possible motion artifacts in rasterizer when using motion steps.
The content of hair placeholder objects was accessed during the scene preprocessing, resulting in unnecessary generation of all hair objects. Fixed.
Fixed black tile artifacts in multihosted rendering of hair.
Fixed missing lightmap baking of secondary bounces of finalgathering with mi_finalgather_store function.
Fixed shading artifacts for shaders using mi_trace_continue function. The returned state->dist value was wrong.
Fixed wrong default importance value for 3rd and higher finalgather diffuse bounces, which could lead to spot artifacts.
In rasterizer, fixed missing initialization of state->dir for the lens shaders.
Improve numerical accuracy of intersections with BSP in scenes with small scene diameter, but a large distance to the camera.
Fixed a precision problem for trimming curve in-out computations for the case of two shared endpoints.
When using images in the 16 bit half-float format with color profiles, the color profiles were not applied correctly. Fixed.
Fixed reading of scalar textures from TIFF images with a single scalar channel. Previously, such textures had to be converted to 3-component color textures.
In debug version only, removed incorrect assert in lightprofile interpolation.
In debug version only, fixed an assert for IES light profiles of type C with symmetry plane at 90-270 degrees and both correction flags (rotate and flip) set.
Platform specific, in debug version only: on 64-bit Windows, removed incorrect assert in mi_lock function.
Platform specific: fixed photon map artifacts for 64-bit Linux platform executable compiled with GCC 4.
Platform specific: fixed geometry merging artifacts on MacOSX X86.
Platform specific: on Linux, fixed a possible mi_openexr.so linkage error if the LD_BIND_NOW environment variable is set.
Platform specific: on 64-bit Linux, fixed possible crash in standalone mental ray when fatal error message was issued.
Platform specific: on AIX, fixed free form surface conversion which resulted in broken tessellations.
Fixed echo of named geometry shaders attached to instances.
Removed possible incorrect error message when echoing requirements for instance groups used as geometry shader arguments.
Fixed echo of some objects with multiple approximations.
For mental ray standalone, fixed the help line for finalgather modes.
Fixed a light linkage compatibility issue in base, physics and paint shaders. If an empty light list argument was passed to a shader, the instance light list was used instead. This is no longer the case, the instance instance list is used only if the mode argument is set to 4.
Fixed a bug in exclusive instance light lists: if such a list contained all lights from the scene, the resulting light iterator looped over the list of all lights, instead of the empty list.
Fixed possible filtering artifacts for the combination of the following conditions:
- the filter is non-trivial (i.e. not box 1 1 ),
- framebuffer(s) always filtered with the default box 1 1 filter are present (for example, the Z-depth framebuffer),
- min samples sampling rate is negative.
Fixed possible artifacts on coarsely tessellated objects in the mib_glossy_reflection shader used with large u_spread and v_spread values. If the reflection goes below the horizon due to the difference between geometry and interpolated normal, always shoot at least one ray above the horizon.
Fixed possible artifacts if more than 2 billion photons were emitted.
For chained lens shaders, propagate state->pri and state->instance properly. In a lens shader, the state->pri can be used to determine if an object has been hit by an eye ray.
For lens shaders in the rasterizer, state->pri refers to the closest primitive hit or is set to null if there was no hit. This allows lens shaders to detect the no-hit case.
For the subsurface scattering shader, fixed low indirect illumination quality problem if the shader is used in combination with finalgather.
In physical_lens_dof shader, fixed a typo leading to too dark images in the finalgather preview mode.
In physical_lens_dof shader, fixed no-hit artifacts when shader is chained with other lens shaders.
Fixed inconsistencies between shader.h and geoshader.h public headers and mental ray functions exposed for the linkage.

Support for OpenEXR

OpenEXR is a flexible image format designed by Industrial Light & Magic, which has been integrated in mental ray 3.3.0.389 and later versions. OpenEXR comes with the following copyright, which applies only to the OpenEXR portion of mental ray:

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
Neither the name of Industrial Light & Magic nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

The ``Pxr24'' compression in OpenEXR comes with the following copyright, which apples only to parts of the OpenEXR portion of mental ray:

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
Neither the name of Pixar Animation Studios nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.

Copyright Information

This document contains proprietary and confidential information of mental images GmbH and is protected under copyright law. The contents of this document may not be disclosed to third parties, translated, copied or duplicated in any form, in whole or in part, without the express written permission of mental images GmbH.

The information contained in this document is subject to change without notice. Neither mental images GmbH nor its employees shall be responsible for incidental or consequential damages resulting from the use of this material or liable for technical or editorial omissions made herein.

mental images®, mental ray®, mental matter®, mental mill®, mental queue™, mental q™, mental world™, mental map™, mental earth™, mental mesh™, mental™, Reality™, RealityServer®, RealityPlayer®, RealityDesigner®, MetaSL™, Meta™, Meta Shading™, Meta Node™, Phenomenon™, Phenomena™, Phenomenon Creator®, Phenomenon Editor™, Phenomill™, Phenograph™, neuray®, iray®, imatter®, Cybernator™, 3D Cybernator™, Shape-By-Shading™, SPM®, NRM™, and rendering imagination visible™ are trademarks or, in some countries, registered trademarks of mental images GmbH, Berlin, Germany.

All other product names mentioned in this document may be trademarks or registered trademarks of their respective owners and are hereby acknowledged.

mental ray Release Notes

Contents

Introduction

Changes by Version

Changes in version 3.7.53.5

Changes in version 3.7.53.4

Changes in version 3.7.53.3

Changes in version 3.7.53.2

Changes in version 3.7.53.1

Changes in version 3.7.53.0

Changes in version 3.7.52.15

Changes in version 3.7.52.14

Changes in version 3.7.52.13

Changes in version 3.7.52.12

Changes in version 3.7.52.10

Changes in version 3.7.52.9.

Changes in version 3.7.52.8

Changes in version 3.7.52.7

Changes in version 3.7.52.5

Changes in version 3.7.52.4

Changes in version 3.7.52.3

Changes in version 3.7.52.2.

Changes in version 3.7.52.0.

Changes in version 3.7.51.0.

Changes in version 3.7.50.15.

Changes in version 3.7.50.14.

Changes in version 3.7.50.13.

Changes in version 3.7.50.12.

Changes in version 3.7.50.11.

Changes in version 3.7.50.10.

Changes in version 3.7.50.9.

Changes in version 3.7.50.8.

Changes in version 3.7.50.7.

Changes in version 3.7.50.6.

Changes in version 3.7.50.5.

Changes in version 3.7.50.4.

Changes in version 3.7.50.3

Changes in version 3.7.50.2

Changes in version 3.7.50.1

Changes in Version 3.7.1.13.

Changes in Version 3.7.1.12.

Changes in Version 3.7.1.11.

Changes in Version 3.7.1.10.

Changes in Version 3.7.1.9.

Changes in Version 3.7.1.8

Changes in Version 3.7.1.7

Changes in Version 3.7.1.6

Changes in Version 3.7.1.5

Changes in Version 3.7.1.4

Changes in Version 3.7.1.3

Changes in Version 3.7.1.2

Changes in Version 3.7.1.1

Changes in Version 3.6.52.10

Changes in Version 3.6.52.8

Changes in Version 3.6.52.7

Changes in Version 3.6.52.6

Changes in Version 3.6.52.5

Changes in Version 3.6.52.4

Changes in Version 3.6.52.3

Changes in Version 3.6.52.2

Changes in Version 3.6.52.1

Changes in Version 3.6.50.21

Changes in Version 3.6.50.20

Changes in Version 3.6.50.19

Changes in Version 3.6.50.18

Changes in Version 3.6.50.17

Changes in Version 3.6.50.16

Changes in Version 3.6.50.14

Changes in Version 3.6.50.13

Changes in Version 3.6.50.12

Changes in Version 3.6.50.11

Changes in Version 3.6.50.10

Changes in Version 3.6.50.9

Changes in Version 3.6.50.8

Changes in Version 3.6.50.7

Changes in Version 3.6.50.6

Changes in Version 3.6.50.5

Changes in Version 3.6.50.4

Changes in Version 3.6.50.3

Changes in Version 3.6.50.2