mental ray Release Notes

Version 3.9.1.37

Feb 4, 2011

Introduction
Changes by Version
Support for OpenEXR

Introduction

These release notes list changes in mental ray since version 3.7.1.1.

Changes by Version

Note: Version 3.9 and later versions of mental ray are binary incompatible with mental ray versions 3.8.x and prior. Partial compatibility with 3.8 is maintained.

Some shaders compiled for previous versions may need recompilation to be used with this version. MetaSL generated shaders compiled for earlier mental ray versions need recompilation.
Version 3.8 cannot be combined with version 3.8.x and earlier versions in network rendering.

Changes in version 3.9.1.37

Bug fixes:

For network rendering, fixed a memory corruption which caused frequent crashes after a slave or satellite was disconnected.
For scenes using builtin BSDFs, fixed possible NaN pixel color artifacts.

Changes in version 3.9.1.36

Bug fixes:

For rasterizer, fixed broken opacity blending if the primary framebuffer is not the first one declared. The issue showed up if a material shader does not call mi_opacity_set(): objects were handled as fully transparent in such case.
Completed the partial fix for rasterizer used in combination with lens shaders in version 3.9.1.34.
Fixed wrong ray type visibility flags used for mi_trace_probe_x() function in combination with BSP2 acceleration.
For raylib integrations using progressive rendering and iray rendering mode, fixed possible freeze of a fast sequence of stop/resume rendering events.

Changes in version 3.9.1.35

Bug fixes:

Fixed wrong gamma correction offset if sRGB (mental ray name: "micsSRGBg") color space was used.
Fixed possible severe shading artifacts for the combination of rasterizer and BSP2 acceleration and motion blur (or motion vectors). State normals are not initialized properly in that case.
For progressive rendering, fixed error estimation code (red channel only was used). This improves sampling distribution and convergency rates for many scenes.
For MetaSL compiler, fixed possible endless loop when compiling call statement.
For MetaSL compiler, fixed possible crash.
For MetaSL compiler, fixed several memory leaks.

Changes in version 3.9.1.34

Bug fixes:

For rasterizer used in combination with lens shaders, fixed possible black image parts or missing lens shader effect (fix in version 3.9.1.34 is incomplete, extended in version 3.9.1.36)
For unified sampling, added a sanity check on min sampling rate. Fixes possible crash.
For iray rendering mode, fixed improper handling of bumped normals for some OEM shaders.

Changes in version 3.9.1.33

Bug fixes:

Fixed possible lighting artifacts if objects (i.e. custom geometry) light sources were used.
For machines with high number of cores, fixed multithreading performance regression in finalgather precomputation stage .
For unified sampling used in combination with rasterizer and low sampling rates (less than 1 sample per pixel), fixed possible black images rendered.
For iray rendering mode:
- For high resolution scenes, fixed possible black stripe artifacts if multiple CUDA devices are used.
- Fixed possible black stripe artifacts and/or crashes due to excessive fragment size used for simple scenes rendered in very large resolution.
- Fixed possible precision artifacts (black lines on the ground plane).
- For visible rectangle area lights, fixed wrong side and size.

Fixes several memory leaks and thread safety issues in MetaSL compiler.
Fixed possible crash at exit or mental ray plugin unload when MetaSL compiler with LLVM backend was used.
For MetaSL compiler, fixed possible unknown tag access fatal when LLVM backend was used to compile shaders with compound (array/structs) input parameters, in particular if the sizes of parameter types in mental ray API do not match the sizes of corresponding types in MetaSL.

Changes in version 3.9.1.32

Feature improvements:

Improved MetaSL compiler reporting (added warnings and notes).

Bug fixes:

Fixed a crash if a DISP socket client (like the imf_disp tool) is detached from mental ray during rendering.
On Linux and Macosx platforms, fixed a performance regression if cached framebuffers were used. A file close/open operation was applied on each tile rendered, which could lead to a significant slowdown in particular when temporary framebuffer files were located on network drives.
For rasterizer used in combination with volume rendering, fixed possible missing volumetric effects and/or black images.
Fixed possible missing shadows artifacts if detail shadow maps were used.
For progressive rendering and iray rendering mode, fixed possible missing effect of output shaders in case when the first framebuffer was not a primary one.
For the debug version, removed assertions triggered when hair objects with wrong bounding boxes were part of the scene.
For map containers, fixed possible issues with the nearest lookups (first element was not the closest one). In particular "search next" functionality could deliver wrong results.
For map containers, fixed missing/wrong reporting of loading ascii map files. Fixed possible failure to load some of such files.
For MetaSL compiler, fixed threat-safety of some compiler operations. Fixed a potential deadlock.
Fixed several issues in the documentation.

Changes in version 3.9.1.31

Feature improvements:

Allow Tiff/OpenEXR/map textures which do not contain MIPMAP pyramids to be cached (highest resolution level is cached, upper levels are created on demand in such case to this is less efficient than having a texture file with MIPMAPs, but more efficient than loading the complete texture into memory).
Improved MetaSL compiler performance. Decreased compiler memory consumption.

Bug fixes:

Fixed possible missing volume rendering effect or black images of rasterizer was used in combination with volume shaders.
For finalgather preview images (-finalgather_display on command line option), fixed color profile correction if the legacy precomputation mode is used. Final rendered images were not affected.
Fixed unified sampling rendering used in combination with scanline on option.
For unified sampling, fixed per-object min/max sampling rates.
For progressive rendering mode, changed adaptive sample generation to match unified sampling.
For interactive progressive rendering and iray rendering mode, fixed rendering of moved lights.
For map containers (including those used for finalgather maps), fixed possible race condition crash when low on memory.
For map containers loaded from the open binary format, fixed possible wrong reporting on the number of elements, and the computation of bounding boxes. The standard 'private' binary format used in particular for finalgather maps is not affected.
For ray-traced particle map objects, fixed a performance inefficiency.
Fixed an error message when assemblies are used in combination with builtin environment lights.
For builtin environment lights, messages and memory consumption is reported in the "IBL" module now.
Fixed possible crash when reading binary map container files with large declaration section.
For scenes with multiply-instanced map objects, fixed possible rendering artifacts and/or crashes.
Fixed several memory leaks in MetaSL compiler kernel.
For MetaSL compiler, added support for geometry_tangent_u/v state member.
For MetaSL compiler, fixed possible infinite loop freeze.
Fixed possible MetaSL compiler crash on deleting a MetaSL shader if nested imports were used.

Shader changes:

In architectural material shader, improved importance-driven area light sampling.

Changes in version 3.9.1.30

Feature improvements:

Allow to use string option with several float values when fewer values are expected. In particular, this allows to use 4 floats for options expecting 3 floats, which is a commit case than an option represents a color.
For MetaSL compiler:
- added support for non-const initialization expressions for shading graphs defined in .msl files
- added support for array typedefs
- added new (Java-style) syntax for arrays.

For imf_disp tool, switched command line options to use underscores for multi-word ones. Improved help banner.
For imf_copy tool, a tiled Tiff file is written out if the -p option is used. Note that the Tiff subformat used in that case is not compatible with mental ray 3.8 and prior, including old imf tools.

Bug fixes:

Fixed possible crash in Irradiance Particles if even number of rays ("irradiance particles rays" string option) was used.
For Tiff image files, fixed several issues in reading and caching various Tiff subformats. Allow scanline Tiff files to be cached. Added a limit to prevent tiny Tiff files from being cached for efficiency reasons.
For raylib integrations: In the progressive interactive mode, fixed a "progressive cannot resume render" error message.
Fixed several memory leaks in MetaSL compiler kernel.
For MetaSL compiler, fixed internal coping of shaders with compound return types.
On Windows platforms, for iray rendering mode, fixed possible wrong device enumeration used in NVAPI test added in version 3.9.1.29. Fixed ECC detection code added in version 3.9.1.29.

Changes in version 3.9.1.29

Feature improvements:

On Windows platforms: iray detects if it is running under SLI or with ECC mode enabled. Both SLI and ECC decrease iray performance significantly, so a warning is printed in such cases. In order to allow this diagnostic, iray is linked against NVAPI library.

Bug fixes:

For unified sampling, fixed sample generation if "samples max" value is set to less than 1.
For progressive rendering mode, changed adaptive sample generation to match unified sampling.
For iray rendering mode:
- Improved driver timeout prevention mechanism in SLI configuration. After a certain number of driver timeout failures the GPU is disabled.
- For procedural OEM textures, scale bump by factor 5. With this change, rendered images look is very close to mental ray's one.
- Added support for OEM double-sided materials.
- Fixed warning messages if standard OEM lens shaders are used in a scene.

For MetaSL compiler, fixed compilation error of phenomena encapsulating light shaders.
For MetaSL compiler, fixed possible crash in parser.
For MetaSL standalone compiler (ray -mslc), removed obsolete command line options.

Changes in version 3.9.1.28

Feature improvements:

For iray rendering mode, report total and available memory for each GPU used.

Bug fixes:

Fixed inefficiency of the kernel small memory block allocator. Some of partially-used pages were not reused for further allocations, which led to a sometimes major memory consumption increase in animation rendering.
For raylib integrations using mi_msg_thread_register()/unregister() functions, fixed possible memory leak of small memory blocks allocated by code in one of mi_msg_thread_register-ed threads and released by other threads.
For mental ray standalone, fixed possible segmentation fault in case of an fatal error (mi_fatal() call).
For iray rendering mode:
- Fixed a crash when rendering an empty scene.
- On multi-GPU machines, fixed possible freeze in case of a falling back to CPU rendering after a failure on one of the GPUs.
- Fixed possible crash for baking of textures via GUI attributes introduced in version 3.9.1.24.
- Fixed possible driver timeout detection failures on CUDA capability 2.0 GPUs.
For MetaSL C++ backend, fixed possible compilation error for bsdfs.
For MetaSL compiler integrated into mental ray standalone (ray -mslc command line option), fixed compiler message reporting.

Changes in version 3.9.1.27

Feature improvements:

For imf_copy tool, added support for writing out MIPMAP pyramids for OpenEXR output image files (-p command line option). With this change, imf_copy -p option is supported for OpenEXR, Tiff and .map files.

Bug fixes:

For texture caching, stripes Tiff files with topdown line numbering (default for Tiff) were read incorrectly. Fixed.

Changes in version 3.9.1.26

Bug fixes:

For the MetaSL standalone compiler (ray -mslc command line), fixed obfuscation option.
For iray rendering mode:
- GUI attribute lookup introduced in version 3.9.1.24 resolves the namespace scope properly
- fixed missing illumination from mirrored (transformation with negative determinant) self-illumination objects
- decrease the default fragment size used to prevent driver timeouts. This is required for scenes with complex OEM shaders.

Changes in version 3.9.1.25

Feature improvements:

For iray rendering mode:
- Improved performance and decrease GPU memory usage
- Added support for a set of OEM procedural textures.

Bug fixes:

Fixed wrong output if the primary framebuffer type is specified to be alpha (a,a_16, a_h, a_fp). The intensity was written out instead.
For unified sampling, if the quality parameter is set to zero, the default quality value is used.
For unified sampling used in combination with rasterizer, the rasterizer is derived from the quality parameter.
For MetaSL compiler, fixed compilation const expression for array size.
For MetaSL compiler, fixed a series of memory leaks.
For MetaSL compiler, fixed possible crash if MetaSL files were not found.

Changes in version 3.9.1.24

Feature improvements:

For mental ray integrations: API GUIs specified inside of shader declarations are now put into the namespace of the shader declaration name. Such attributes are now echoed together with shader declaration. mental ray echo options have been extended with a flag (gui) allowing to echo all gui attributes from the global namespace (i.e. not defined inside of the shader or phenomena declarations). This options is disabled by default. An equivalent mental ray standalone option (-echo gui <out_filename>) has been added.
For iray rendering mode, allow baking of 3rd-party textures if the resolution gui attributes ("bake_resolution_u", "bake_resolution_v", "bake_scale_u", "bake_scale_v", "bake_coord_shader") are specified.
For rendering using unified sampling, allow diagnostic framebuffers to be specified explicitly: buffers matching names "mr_diagnostic_buffer_*" are used for corresponding diagnostics. In particular, this allows to specify image file names and types where the diagnostic data is saved.
For MetaSL LLVM backend, improved runtime performance of generated shaders by allowing LLVM compiler optimization. The optimization can be disable with the registry " value "0" end registry"{_MI_REG_METASL_LLVM_OPTIMIZATION_LEVEL}" value specified to "0".
For imf_disp tool, allow to switch between images with Ctrl Left/Right keys.
For imf_comp tool, added a new command line option -u. As mental ray creates pre-multiplied rgb/a images by default, imf_comp composites in premultiplied mode, unless the new u-u option is used.

Bug fixes:

Fixed possible sporadic crashes in finalgather rendering. The crashes were observed on some multicore machines on 64-bit Windows platform.
Fixed possible rare crash in the kernel memory manager.
Fixed a crash when -reload command line option (auto-converting of objects into placeholders) was used. The -reload option is still incompatible with gzipped .mi scene files.
For BSP2 acceleration, avoid a crash if a shader traces a ray from a NaN origin or in a NaN direction.
For BSP2 acceleration used to render scenes with motion blur, avoid possible unneeded object tessellation.
On Linux platforms: in iray rendering mode, fixed possible freezes when multiple CUDA capability 1.3 devices were used.
For MetaSL compiler, fixed compilation of string-from-double constructor.
For MetaSL compiler, fixed support of Spectra values in XMSL files.
For MetaSL LLVM backend, fixed possible shader compilation failure on 64-bit Windows platforms.
For imf_disp tool, fixed possible freeze if an attempt to load an image of an unknown sub-format is done.

Shader changes:

For base mib_map_get_* shaders used for map object data retrieval, fixed the return types in .mi declarations.

Changes in version 3.9.1.23

Feature improvements:

Improved performance for motion blur scenes with large number of fast moving objects when rendered with BSP2 acceleration.
For imf_copy tool, added support for writing out MIPMAP pyramids for TIFF output image files (-p command line option).
Cached texture statistics are printed out sorted by number of accesses.

Bug fixes:

Fixed possible memory corruption in MetaSL compiler frontend and LLVM backend.

Changes in version 3.9.1.22

Bug fixes:

Fixed mental ray kernel's internal memory management failure to re-claim pages with small allocated memory blocks after working thread shutdown. This failure may be perceived as memory leak in long animation rendering.
Force writable textures to be unfiltered. Update of pyramids on texture modifications is not supported.
For map containers fixed:
- search path substitution not applied properly
- minor memory leaks in map container declaration, definition headers and global values
- null tag access fatal when echoing a map object with map which failed to load.

Shader changes:

For production mip_motionblur()and mip_motion_vector() shaders, fixed possible crash when use_coverage parameter was set.

Changes in version 3.9.1.19

Feature improvements:

For MetaSL compiler, added support for shader graphs and phenomena specified in .msl file.

Bug fixes:

Fixed a crash when empty scene is rendered with BSP2 acceleration and motion blur enabled.
Fixed a crash if a TIFF image file with (exactly) two channels is loaded. The first channel is used in such cases, the second one is ignored.
Fixed a memory leak when map cache is flushed due to mental ray running low on memory.
For ccmesh objects without texture spaces, fixed possible crash if mesh contains polygons with 5 or more certifies.
On Linux and MacOSX, fixed possible crash when loading ASCII map files.
For MetaSL shaders, fixed raster space transformations for orthographic camera.
Fixed doxygen comments in public header files.

Changes in version 3.9.1.18

Feature improvements:

Added import of map files in ASCII format. The files follow the general .mi file grammar, but allow only map declaration and definition elements, as well as nested $include statements.
Per-texture cache statistics are printed out only if IMG module file verbosity is set (-message img files -- command line option). Combined all cached textures and cached framebuffers statistics is printed after each frame (if not void).
If mi_rc_run()'s mode miRENDER_ACCEL_DELETE is set, map container cache is flushed from memory. In particular, this applies to the memory occupied by finalgather map from the last frame. Map container cash is also (partially) flushed if mental ray is running low on memory.
For iray rendering mode:
- Report the memory available on each CUDA device
- Improved performance on CUDA 2.0 ("Fermi") capable devices.
- Improved convergency rate for scenes with many lights and objects.

Bug fixes:

Fixed a fatal/crash in mi::integration::Access/Edit_result_image when the call was done after the scene postprocessing.
For mental ray integration, expose functions mi_colorprofile_to/from_internal(). Some OEM integrations depends on these function used to convert colors from / to internal color space, in particular in DISP callbacks.
Fixed minor memory leaks in BSP2 acceleration.
For iray rendering mode:
- Fixed possible freezes or crashes when rendering images of high resolution not fitting into the GPU memory.
- Fixed possible failure to detect CUDA kernel execution error. In particular,this fixes stripe artifacts when GPU + CPU mixed rendering is used.
- Fixed difference in the images rendered on CPUs and GPUs.
- Fixed possible pattern artifacts in textures using Perlin noise.
- Fixed possible infinite loop in light sampling for scenes with transparent materials.
- Fixed artifacts for Normals_height_bumpmap MetaSL shader.
- Added support for shaders returning alpha value of their input.
- Fixed a warning when a light falloff exponent was used instead of the value provided by a light shader.
- Fixed possible environment sampling artifacts.
- Fixed conversions of phenomena attached to lights.
- Improved support for OEM shaders.

Shader changes:

For mip_bump_basis shader, fixed the basic computation for non-orthonormal transformations of geometry instances.

Changes in version 3.9.1.17

Feature improvements:

For the dsm_copy (detailed shadow map inspector), added an option -l. If used, the Z depth is stored to the color image alpha channel. Note that many image viewers will not display such output images in a correct way.

Bug fixes:

For raylib integrations using iray and progressive rendering in interactive mode, fixed possible freeze in a fast sequence of stop and resume events.
Fixed a database tag and memory leak when animation was rendered with finalgathering and finalgather rebuild on option.
Some of memory blocks were not accounted into the memory statistics, fixed.

Changes in version 3.9.1.16

Bug fixes:

Fixed possible crash for ray-traced hair rendering with motion blur.
Platform specific: for 32-bit windows executables (in particular, running on 64-bit windows systems), fixed segmentation fault crash in ray-traced hair rendering.
If unified sampling is used, fixed possible artifacts in user framebuffers. In earlier versions, if no value was set by a shader for a raster sample, the previous sample's value was used due to a missing initialization.

Shader changes:

For subsurface scattering shader, fixed possible black speckles self-shadowing artifacts.

Changes in version 3.9.1.15

Feature improvements:

For cached TIFF textures, added support for MIPMAP pyramids in the TIFF files.
For MetaSL language, added set_importance(float) for Occlusion_options, Irradiance_options, Trace_options without warnings.

Bug fixes:

Cached TIFF textures were displayed topdown, fixed.
When rendering finalgather with "finalgather rebuild on" option, a finalgather map created with mental ray 3.8 and prior was not updated with the new content, fixed.
Fixed possible crash in MetaSL compiler when compiling XMSL files.

Changes in version 3.9.1.14

Feature improvements:

For unified sampling, added the time spent for each pixel buffer ("mr_diagnostic_buffer_time"). The time stored is only approximate any may vary significantly for individual pixel, in particular when a ray used to compute pixel triggers tessellation of geometry, loading of texture or assemblies, executes init shader etc. Few pixels having significantly higher rendering times are expected for all non-trivial scenes.

Bug fixes:

Fixed unnecessary resending of tiles through DISP protocols and callbacks.
On gcc4-compiled platforms fixed wrong texture cache reporting due to gcc4 compiler optimization bug.

Changes in version 3.9.1.13

Feature improvements:

For map containers, added API (Map_base::Map_status_write() parameter) to specify in the file is saved in optimized binary, simplified binary or ascii format.
For MetaSL phenomena, added support for external/internal input/output parameter name mapping. mental ray API and .mi syntax operate on external names. Runtime MetaSL cross-shader calls and message passing use internal names.
For MetaSL shaders, improved color texture lookup quality when derivatives are provided. Elliptic filtering lookup is used in that case now.

Bug fixes:

Fixed possible out-of-memory fatal in rare cases when a memory block was mi_mem_reallocated to a zero size. In particular, this was observed in rendering animations with irradiance particles.
Texture cache reporting is switched to prints stats after each frame rendered. Fixed an issue with SPM-protected mental ray standalone, which exited before the statistics were printed out in version 3.9.1.10.
On Linux/Macosx systems, fixed possible open file descriptor leak if both texture caching and cached framebuffers were used.
For raylib integrations: with unified sampling, 4-channel buffers were used for 1-channel (s or a) image types. Fixed.
Do not print an error message if output shader or raylib integration code attempts to access a framebuffer with out-of-range index. A null image pointer is returned, like in earlier mental ray versions.
For MetaSL LLVM backend, fixed ray type comparison operators and shader type state variable.

Shader changes:

For the subsurface scattering skin shader, fixed possible stream of error messages for scenes using hull materials (those with no surface and bsdf shaders).

Changes in version 3.9.1.11

Feature improvements:

iray rendering mode is switched to use CUDA 3.2. This may require a graphic driver update on machines using iray. This improves iray performance and fixes broken rendering on Fermi (GTX 400 series) cards.
Improved performance of TIFF texture caching introduces in version 3.9.1.10 (reduced the number of IO operations).
For raylib integrations, added mi::integration::Access/Edit_result_image::get_result_image_count() static method to get the number of images. For stereoscopic rendering the returned value is the number of images per camera.

Bug fixes:

Fixed a performance regression in a special case when rasterizer with motion steps 1 was used in combination with BSP2 ray tracing. In this case, a motion version of the BSP2 tree was constructed, though only a static one is used.
Fixed echo of verbatim map containers.
For hair rendering, rendering abort test was missing, which could lead to a long reaction time. Fixed.
For MetaSL compiler, fixed several issues for swizzles.
Fixed an error message printed with -iray off command line options is used.

Shader changes:

Added a new shader library userdata.dll/.so/.mi. This library contains a set of utility mib_data_* shaders for accessing named user data values attached to the scene graph elements. The documentation is provided in the userdata.html manual file.

Changes in version 3.9.1.10

Feature improvements:

Enable texture caching for TIFF file format. In this version, MIPMAP pyramids form TIFF files are not supported yet. Non-RGBA(A) images, images with strip height equal to 1 and with components count equal to 1 are not cached.
Improved texture cache stat reporting. If a texture is (re)loaded several time, the consolidated statistics including the number of re-loads is printed at the end of process / before raylib shutdown. Number of (re)loads of texture is added. Stats clearly distinguish between cached textures and cached framebuffers.
Performance of the hair shadow intersection code has been improved.
For raylib integrations using progressive rendering, added an enum member MI_RC_PROGRESSIVE_FG_PRECOMP. This value is returned by mi_rc_progressive_*() functions during finalgather precomputation stage.
Added registry {_MI_REG_MAP_CACHE_SIZE} which allows to specify the memory size used by map containers. Value is interpreted in MBs. This is an addition to already existed command line option -map_cache_size N.
For MetaSL compiler, check grammar stricter to catch the following illegal constructs:
- input, output, member sections in structs
- typedef, enum, constructor, destructor, or method declaration in struct
- output section in bsdf.

For the dsm_copy tool, fixed the 'color combine' mode. Added support for conversion to 'z' type images.

Bug fixes:

Fixed a crash in mi_shaderstate_set() function, if an existing value of a key is reset to a new value of a larger size.
For ray integrations based on static raylib library: compile the PNG library with C++ and wrap it by MI namespace. This solves possible symbol conflict if the integrating application uses its own copy of PNG library.
For progressive rendering and iray rendering mode, fixed possible freeze if stop and resume events were issued with low time interval between them.
Fixed possible wrong min/max samples warning if unified sampling was used.
For MetaSL shader, ignore metasl_connected_normal parameter if it is not used as an attachment (this is a reserved parameter name generated for MetaSL shaders for attaching bumped normals).
For MetaSL compiler, fixed a serious of bugs related to shadowing elements defined in an outer scope by elements with the same name in the inner scope.
For imf_disp tool, fixed image format error reporting.

Shader changes:

For subsurface scattering shader, unroll the changes in version 3.9.0.13 due to the possible artifacts in reflection. Instead, a new parameter "fallback_shader" as added to the misss_fast_shader(_x) shaders. This parameter allows to make sure that the proper light is used if lightmap samples are completely missed due to small lookup radius. Old version of the shaders used lambert illumination. If this parameter is not specified, the old behavior (lambert) is maintained.

Changes in version 3.9.1.7

New features:

Added two new utilities for usage in production rendering:
- dsm_copy allows to create a slice visualization image of detailed shadow maps, which can be used for tuning the quality of shadow maps. Usage:
  dsm_copy/dsm_copy [-h] [-c] [-v] [-z float]] detailshadowmap [outimage [outformat [outtype]]]
  
  -h option print online help
  
  -c combine z and color/alpha in single image
  
  -v prints verbose info about the detailed shadow map
  
  -z <positive float> specify the z-depth threshold to sample values from.
- imf_comp is a simple compositing tools which can be used to combine several images rendered with camera window (-window command line option) into a single one. Usage:
  imf_comp [-h] [-v] -o imagefile [compfile...]
  
  -h option prints online help
  
  -v prints verbose output
  
  -o options specified the file name of the composites image.

Feature improvements:

In case of finalgather precomputation, create rendering framebuffers after the finalgather map is build. In case of in-memory framebuffers this reduces memory consumption during finalgather precomputation phase.

Bug fixes:

For rendering with finalgather, fixed possible error message on unpinning null tag. Regression in version 3.9.1.4.
Fixed possible crash in rendering finalgather in progressive interactive mode.
For hair rendering, fixed possible shadow artifacts due to a numerical precision issue.
For MetaSL compiler, fixed compiler error on operator= for enumerations.
For shader writers: fixed a compiler warning on 64-bit Windows platform in shader.h include file.

Shader changes:

For experimental rendering map particles, added the new base shaders added in version 3.9.1.1 are renamed to mib_map_get_scalar/integer/vector. Added shaders mi_map_get_transform/scalar_ array/integer_array for retrieving transforms and arrays. For arrays, the complete array is returned and the caller must assure that the size of the buffer is sufficient.
For mib_continue and for rendered particle map data access shaders (mib_map_*), added missing "apply texture" qualifier to the declaration.

Changes in version 3.9.1.5

Bug fixes:

For progressive rendering, fixed possible crash when a camera is deleted or incrementally changed. Regression in version 3.9.1.4.
Fixed possible numerical precision artifacts in hair shadows.

Changes in version 3.9.1.4

Feature improvements:

For interactive progressive rendering, improved the granularity of detected scene changes. Updating light shaders, updating light instance transforms, updating lens shader and updating environment shader are now detected individually.
Reduced memory consumption of the BSP2 acceleration for motion blur scenes in a special case when no rasterizer, shadowmaps, are scanline used, and shadows are limited to the simple mode (not segmented or sorted).

Bug fixes:

Fixed confusing warning message on the progressive max samples.
Fixed possible crash (frequent on Windows platforms) in framebuffer image output when contour shaders were used.
For multihosted rendering, fixed possible crash or missing indirect illumination in rendering sequences of several frames.
For integrations, fixed mi::integration::Access/Edit_result_image classes for the finalgather preview framebuffers created in finalgather precomputation stage.
In multi-hosted rendering, fixed a crash of the slave/satellite hosts at shutdown.

Changes in version 3.9.1.3

Feature improvements:

Improved controllability and reporting of texture caching.
- Added a command line option -texture_cache_size <MB> and registry {_MI_REG_TEXTURE_CACHE_SIZE} . If set, the value is taken as a limit of texture cache in MBs. By default (value 0), the dynamic texture cache limit based on the size of all textures loaded to mental ray is used (as in mental ray 3.8 and prior). The minimal value of the cache is 20 MB. Command line option takes precedence over the registry.
- Added a command line option -texture_cache <mode> and registry {_MI_REG_TEXTURE_CACHE}. The value can be one of "on", "off" or "local". The default value is "local", meaning that only local (and .map, which are automatically local) textures can be cached. This is the default behavior. With "off", texture caching is disabled and textures are loaded/memory-mapped completely. With "on", the local texture modifier is ignored and all textures with image formats supporting caching (currently .map, tiled .exr, .iff, .bmp) are cached. This mode should not be used with multi-hosted rendering. Applications integrating raylib should make sure that all non-local textures are deleted before this mode is enabled.
- Print out texture cache mode and cache size limit in the options info banner before each frame rendering.
- Print out the statistics on the number of pixel accesses and per-thread/local cache coherency rate for each cached image file (i.e., for cached textures and cached framebuffers). Print out accumulated statistics before process/raylib exit.
For TIFF images, utilize the full libtiff library. In particular, reading of LZW-encoded images is supported.
For raylib integrations, added support for a combination of finalgather preview mode with progressive interactive rendering utilizing application framebuffer.
For MetaSL compiler, allow typedefs inside of shaders and functions.
For MetaSL compiler, improved static code analysis.
For MetaSL LLVM backend, added support for techniques.

Bug fixes:

Fixed a crash in progressive rendering stopped on the quality criteria reached (regression on version 3.9.1.2).
Fixed a crash when finalgather previewing mode (-finalgather_display on command line option) was used in combination with unified sampling.
Fixed several memory leaks in MetaSL compiler.

Changes in version 3.9.1.2

Feature improvements:

Improved integration workflow for mental ray progressive rendering and iray rendering mode:
- Added support for incremental instance updates (materials and transformations).
- mental ray kernel monitors database tags for modifications (with mi_scene_edit/growedit functions), mi_api_* functions). This information is used on avoid unnecessary re-baking in iray rendering mode.
- Tags deleted by integrating application during progressive rendering are not destroyed immediately. Instead, they are accumulated in a pool which is cleaned up when the rendering is finished.
For iray rendering mode:
- added support for texture clipping and alpha
- reduce memory consumption by detecting duplicated textures
- in interactive mode, added support for changes of the environment shader
- added statistics for texture memory
- fixed black pixel artifacts
- improved support for OEM shaders
For displacement shaders changing miState::point (displacement not in the direction of normals), print a warning if the sum of the point shift and displacement along the normal is larger than maxdisplace limit. In such case, reduce the displacement by normalizing the sum of two components to the vector of maxdisplace length.
Use mental ray memory management for blocks allocated by the integrated OpenEXR library.

Bug fixes:

For displacement shaders called on the triangle meshes in the presampling face, miState::normal was not normalized. Fixed.
For displacement shaders called on ccmesh objects, the modification of miState::point by a shader was ignored. Fixed.
Fixed possible rare race-condition crash during finalgather precomputation, more likely for simple scenes.
Fixed possible assertion/crash when finalgather precomputation with legacy mode ("finalgather precomputation legacy" string option) was aborted.
For progressive rendering, fixed tone mapping artifacts if no filter was applied.
For MetaSL shaders, fixed shader parameter initialization.
For MetaSL shaders, fixed missing textures which were not exposed as input parameters of phenomena.
On MacOSX platforms, fixed a memory corruption in STL file streams. In particular, this caused crashes if OpenEXR image files were used. Regression in version 3.9.1.0.

Changes in version 3.9.1.1

Feature improvements:

For iray rendering mode, incremental updates for lights, geometry, textures and materials are now supported.

Bug fixes:

For raylib integrations: for Access_result_image/Edit_result_image() calls, fixed possible deadlock if more than one image is accessed simultaneously.

Shader changes:

For experimental rendering map particles, added new shaders mib_map_integer_picker(), mib_map_scalar_picker(), mib_map_vector_picker() and mib_map_color_picker(). This shaders take a map filed name as a parameter. The result is set to and the value of the particle currently being hit. If object is not map, or if the map does not contain corresponding value, shaders return miFALSE.

Changes in version 3.9.1.0

Feature improvements:

For MetaSL compiler:
- Added trace() function without explicit differentials. The differentials are computed by the kernel in this case.
- Empty structs are allowed in MetaSL language now.
- Compiler kernel has been refactored for better performance. A series of memory leaks has been fixed.
Hair rendering uses the new acceleration structure now, which result in significantly better performance and lower memory consumption. For backward compatibility / performance comparison, the string option "hair acceleration" could be set to off.
Integrated OpenEXR library is updated to version 1.7.0.

Bug fixes:

Fixed possible artifacts and crashes with scenes containing several renderable map objects.
For MetaSL frontend, fixed handling of empty annotations.
For memory management, memory allocated for the management system itself was not counted in the statistics. Fixed.
For geometry shaders and raylib integrations: the inheritance information on the leaf instances was broken. In particular, the miLEAFHISTORY macro did not work properly. Fixed.
Fixed a regression in version 3.9.0.14. If iray rendering mode was used with "gpu" mode (default), and no CUDA-capable GPUs were found in the system, mental ray exited with fatal. Fixed similar regression if echo iray mode (-echo iray command line option) was used, independently of the iray mode specified.

Changes in version 3.9.0.14

Feature improvements:

For texture caching, the dynamically computed texture cache memory is based on the in-memory size of the uncompressed textures instead of the file size used in the earlier version. In particular, the memory used for rendering no longer depends on the type of texture compression used.
For particle map containers, improved performance of merging operations.

Bug fixes:

The value for "primary" qualifier of a framebuffer was ignored. Now interpret the "primary" off flag correctly.
For raylib integrations, fixed a crash when mi::shader::Framebuffer::reset() method was called.
For unified sampling feature, fixed writing out of multiple framebuffers.
For unified sampling feature, fixed finalgather diagnostics.
Fixed an assertion in memory management system. Possible memory leak at shutdown messages are downgraded to debug.
Fixed echo of trilist object with some rarely used vertex layouts.
Fixed a crash in single-threaded baking of environment map.
For unified sampling, fixed possible NaN pixel value artifacts on the tile borders.
Fixed possible assertion for ccmesh objects.
For iray rendering mode, fixed a fatal for 2D texture transformation MetaSL nodes.

Changes in version 3.9.0.13

Feature improvements:

For iray rendering mode, added a string options "iray mode" which controls the selection of GPUs vs. CPUs. The valid values are:
- "gpu". Only GPUs are used. CPUs are used if the machine has no CUDA-capable GPUs available.
- "cpu". Only CPUs are used.
- "mixed". Both GPUs and CPUs are used.
The default value is "gpu".

Bug fixes:

Fixed a crash in ccmesh/sds displacement if displacement shader returned a color (scalar value is expected).
Internal mi_img_mmap_address function is removed from the public headers.

Shader changes:

For subsurface scatter shader, fixed illumination when no samples could be found within the specified lookup radius.

Changes in version 3.9.0.12

Feature improvements:

Drastically improved the performance of the MetaSL LLVM backend.
Added final gathering diagnostics to unified sampling modes.
In approximate mode, the number of light samples for the environment lighting is calculated based on the value of miState::importance.
Added support for a new map file format (open binary), read mode only (open binary map files can be read in, but not written out as of now).
For the experimental ray-traced particle maps, added two new mi_query() modes, miQ_PRI_PARTICLE_TYPE (that returns 0 for spheres, 1 for tubes and -1 in case it's not a map object) and miQ_PRI_PARTICLE_UV (returns the UVs of the intersection point on the particle).

Bug fixes:

Fixed an issue with miState::derivs for freeform-surface displacement presampling.
Fixed an issue with hair radii - sometimes they were not properly accounted for, causing artifacts.
Fixed a bug in imf_disp that prevented to stop an animation render after the first frame.
Fixed an issue regarding an error message being printed before adding a framebuffer in case there was none. Now only an info message is printed.

Changes in version 3.9.0.11

Feature improvements:

For imf_disp, added the build number to the "About" dialog.
Reduced the blur in texture lookups, differential-based filtering was too strong with the default settings.
Improved the map caching system by making it more adaptive and more responsive to changing memory conditions.
For iray rendering mode, added a new mode (ground) to the environment dome modes.

Bug fixes:

Fixed an issue with mi_raylib_license_get returning the wrong number of licenses.
Fixed an issue with missing derivatives for displacement presampling.
Fixed an issue with filtering, no filtering is applied for secondary rays when differentials are not enabled.
For iray rendering mode, fixed the echoing of spot lights.

Changes in version 3.9.0.10

Bug fixes:

Fixed a crash in imf_disp that occurred when opening an image with only one layer after displaying a different layer in another image.

Changes in version 3.9.0.9

On MacOSX, minimal supported version is increased to 10.5 (Leopard).

Feature improvements:

For BSP2 acceleration used with motion transformation blur in scenes with multiple instancing,completely removed per-triangle data not shared among instances.
Fixed motion interpolation if multistep motion is used.

Known limitation in this version: possible shadow artifacts if segment shadow mode is used.
For unified sampling, added a possibility to specify less than 1 samples per pixel with "samples min" string option in the range [0, 1].
For unified sampling, improved the adaptivity criteria.
Added two new mi_query() modes which fill the gap in kernel functionality provided to (geometry) shaders:
- The miQ_HAIR_NUM_USERS mode gives the number of per hair strain user data attached.
- The miQ_HAIR_USERS mode fills the buffer with a copy of the user data content.
- The miQ_HAIR_NUM_TEXTURES mode gives the number of per hair strain texture spaces attached. (The miQ_NUM_TEXTURES mode result is the sum of numbers of per hair strain and per vertex texture spaces).
Added new mi_query() mode miQ_PRI_NUM_VERTICES. The result of the query mode would be the number of vertices of the primitive hit. In the current version, the mode will return false except for triangles, and for triangles the returned value is 3. Base shaders source code examples are extended to use this query mode for safety reasons. Support for hair is planned for later mental ray 3.9 version.
Added new mi_query() mode miQ_NUM_MOTIONS, which delivers the number of motion vectors after the tessellation. The miQ_PRI_VERTICES_MOTION query mode introduced in version 3.9.0.6 now requires an additional parameter which selects the motion vector index. This fixes the functionality gap: with deprecated mi_tri_vector() function it was possible to obtain the first motion vector only.
For particle maps, improved performance for large particle maps when only a fraction of the data resides in memory.
For the experimental ray-traced particle maps, added a new mi_query() mode miQ_PRI_PARTICLE_INDEX. The query result is the index of a particle in the map.

Bug fixes:

For echo, added proper escaping of the quote sign (") in API names and string values.
Fixed possible missing shadow artifacts if classic (not detail) shadow maps with no shadowmap bias were used for scenes utilizing shadow cast and receive flags (i.e., some object having just one of the flags set).
Fixed wrong motion put to motion blur buffer if a shader on a moving object calls mi_trace_continue() function, and the object hit behind is static. In particular, this is the case for the mip_card_opacity production shader.
Fixed an error in computation of motion factor ("geom displace motion factor" string option) for ccmeshes (since version 3.9.0.6). For slow motion, the tessellation sometimes was coarser than for equivalent static case.
For the deprecated mi_tri_vector() function called from a displacement shader, the value for the first vertex were returned for all vertices. Fixed. Regression in mental ray 3.9.
For MetaSL compiler, fixed several memory leaks.
For MetaSL compiler, fixed shader inputs with external names.
For MetaSL LLVM backend, added missing support for String constructors.

Shader changes:

For the production mip_rayswitch_advanced shader, fixed a crash if an attached parameter shader had more than one output. The first output of the latter shader is taken as a result of mip_rayswitch_advanced now.
For mia_physicalsky shader, fixed sun disk showing at night rendering artifacts.

Changes in version 3.9.0.8

New features:

Experimental / early work in progress: particle maps as native primitives.
mental ray syntax has been extended, allowing an object to contain particle maps. Unlike "named" particle maps which are used by shaders, those maps become a part of the scene DAG and rendered by mental ray kernel.

Extended .mi syntax with:
```
    object
        ...
        particle "spheres" | "tubes"
        [ "take_what" "from_this_map_field"]
        map
            [ map file list ] |
            "<map declaration name>" [ map elements ]
        end map
    end object
 
```
Currently, two particle element types are supported: spheres and tubes. More types may be added in the future. For integrations, added function mi_api_map_obj_type() which allows to specify a type of the map.

With the next syntax statement, it is possible to specify which map files should be used as special parameters. For examples, for spheres "radius" "my_rad" would instruct mental ray to interpret the value "my_rad" of a particle as a radius. For spheres, "radius" is the only name supported. For tubes, it is possible to specify "radius0", "radius1", "direction" and "length" in addition. For integrations, added function mi_api_map_obj_field().

The body of the map is similar to the declaration of the named map object, with the exception that the name of the map is omitted.

For integration: the miObject_type enum has been extended with miOBJECT_MAP entry.

In this alpha version, particle maps have many limitations including o supportno support for motion blur, rasterizer, shadow maps, complex shaders.

Feature improvements:

In unified sampling (introduced in version 3.9.0.6), changed "samples min" / "samples max" options to float. This is done as work in progress on allowing less than 1 sample per pixel.
For BSP2 acceleration used with motion blur in scenes with multiple instancing,reduced the amount of data not shared among instances by approximately one third.
Initial implementation of unified sampling in rasterizer mode has been enabled. Due to the fundamentals of the rasterizer algorithm, the semantic of parameters is slightly different. The raster sampling pattern is taken based on the max samples parameter. The min samples parameter is used as the minimal number of shading sampler per pixel. The quality parameter increases the number of shading samples.
For finalgather precomputation, "fake" frames no longer inserted by the talk protocol for finalgather precomputation.
Added new builtin environment shader
```
    declare shader
        color "builtin_environment_lookup" (
            color texture "tex"
        )
        version 1
    end declare
```
If the shader is used in the scene and the IBL settings are compatible (IBL scale set is not set or set to 1.0, no IBL resolution set or the resolution matches the resolution of the builtin shader texture parameter), when the texture is applied directly as a spherical lookup, bypassing internal environment map re-baking.

This shader is in particular intended to be used in iray rendering mode, there environment precomputation may consume a significant fraction of rendering time otherwise.
For ray traced hairs, reduce the memory used by acceleration structures.

Bug fixes:

Fixed a crash when deprecated mi_tri_vector() function was used to retrieve bumps.
Fixed sporadic memory and scheduling manager crashes on multi-CPU machines.
In the rasterizer, fixed darker visible environment.
For cached textures (some OpenEXR, Iff, BMP and .map files), color space conversion was not applied. Fixed.
The .map textures which require color space conversion cannot be memory mapped. Instead, texture caching is used now.
Added command line option -iray gpu (in addition to -iray on/cpu). If set, iray rendering is done on GPUs only.

Changes affecting compatibility:

The mi_msg_hyper() function has been removed from the integration interface. In many cases, in particular for virtual machines, there is no way to reliably determine if the machine is running hyperthreaded CPUs.

Changes in version 3.9.0.7

Feature improvements:

Print info banner on the sampling parameters if unified sampling introduced in version 3.9.0.6 is used.
Improved documentation.

Bug fixes:

Fixed a crash when rendering with zero framebuffers was started.
Fixed importance sampling for BSDF shaders glossy component.
Fixed several minor issues in MetaSL compiler.

Changes in version 3.9.0.6

New features:

New sampling scheme.
In this version, the sampling need to be enabled explicitly with string option

"unified sampling" on

The controlling string options are:
- "samples min" <int> (<float> since version 3.9.0.8). Specifies the minimal number of samples per pixel. Unlike in the old scheme, this value is linear, not exponential. The minimal acceptable value is 1.
- "samples max" <int> (<float> since version 3.9.0.8). Specifies the maximal number of sampler per pixel before stopping the rendering.
- "samples quality" <color> Synthetic criterion to stop rendering. Higher value lead to better quality. The "soft" valid range for color components is [0,1], however values larger than 1 are accepted. The default value is 0.5.
- "samples error cutoff" <color> This is the error level where further improvements are considered negligible. The default value is zero (black). Setting the value higher leads to less samples taken in the dark area of the image.
- "samples per object" <bool> If enabled, per-object min/max sample values are taken into account. Note that the semantic of this values is different from the old sampling scheme (linear instead of exponential).
New sampling scheme provides per-color framebuffer control over the filtering. The framebuffer .mi syntax is extended with:
- filtering <bool> Enables / disables filtering. If framebuffer filter is not specified, the one from miOptions is used.
- filtering [clip] "filter name" [support_x] [support y] Specifies the filter type ("box", "gauss", "mitchell", "lanczos"), clipping version (only relevant for mitchell and lanczos), and a support size (floating value in pixels).
- clip <bool> This controls clipping of samples before pixel value is computed. By default, this is on if the framebuffer is not of float/half resolution.
For the integration API, the Access_fb::set() method recognizes the "filter type", "filter support", "filter support x", "filter support y", "clip filter" and "clip" attributes.
miOptions::images_info is deprecated.

miOptions::no_buffers gives the actual number of framebuffers used in the rendering process. Note that this number may be different from the number of framebuffers explicitly specified in the scene.

This feature is WIP.
For output shaders, added classes Access_out_img and Edit_out_img (shader_output.h), which allow to access framebuffer images by name and by index.
The enumeration miRC_IMAGE_* is deprecated.

Feature improvements:

Improved performance of the BSP2 accelerations in motion blur scenes. Reduces memory consumption for scenes with multiple instances. Improved accuracy of motion blur images by shading with time-accurate geometry normal, interpolated normals and shadow tolerances.
For ccmesh objects, implemented support for motion-driven displacement quality reduction enabled with "geom displace motion factor".
Improved performance of ray-traced hairs.
For ray traced hair, mental ray adds caps at the end of the hair cylinders. Added a string option "hair caps" <bool>, to make caps optional. By default, caps are enabled.
Finalgather maps are switched to the MAP container. The advantage is that for large maps it is no longer require to keep complete finalgather map in memory as the map container caching system is utilized. Also, when a precomputed finalgather map is loaded from file, the load operation is significantly faster. New fg_copy utility can be used to convert mental ray 3.8 and prior finalgather maps into the new format and vice versa.
For the stereoscopic rendering offaxis mode, added a possibility to specify parallax distance different form the camera focal plane.
The .mi camera syntax is extended with stereo offaxis <eye_separation> [<parallax_distance>] statement (the last parameter is new and optional). The default value for parallax distance is camera focal distance.

For raylib integrations, miCamera::parallax_distance is added to the camera structure.
For stereoscopic rendering, change the scheme used for output files. The registries {_MI_REG_STEREO_LEFT_SUFFIX} and {_MI_REG_STEREO_RIGHT_SUFFIX} can be used to specify the suffix added to the file name before the extension(s) (if any). The default values are "_left" and "_right". (For example, if no registers are specified and a framebuffer file name is output.my.exr, when the files output_left.my.exr and output_right.my.exr will be written.) Note that this naming scheme is incompatible with the fixed prefixes lgt_ and rgt_ used with earlier mental ray versions.
Performance and robustness of MetaSL compiler has been improved.
Improved support of MetaSL phenomena in mental ray, including material phenomena. Enable ior() node.
Improve compatibility of the native C/C++ shader packages with MetaSL shaders.
When the string option "light relative scale" <float> or the registry "{_MI_REG_LIGHT_RELATIVE_SCALE}" is set, the colors passed from MetaSL light shaders to native surface shaders are multiplied by the specified value. The colors passed from native C/C++ light shaders to MetaSL surface shaders are multiplied by the inverse of the value.

Some of the OEM native shader packages use non-standard conventions. By using the string option/register value of 1/Pi, identical/consistent rendering results may be achieved with scenes using both native and MetaSL shaders, as well as builtin BSDFs.
For MetaSL shaders, do not load MetaSL LLVM backend library (gen_llvm.dll/.so) if MetaSL C++ backend is used. Improved the messages: an error message is printed if gen_llvm.dll/.so cannot be found. An info message containing the full path is printed when the library is loaded. The path list search is downgraded from info to debug. The LLVM backend is not loaded when mental ray standalone is used as MetaSL compiler (ray -mslc command line).
Job scheduler: improved multi-threaded and multi-hosting scalability. Reduced number of hosts synchronizations in multi-hosted rendering.
Memory management: improved multi-threaded scalability. Frequency of locked operations is reduced significantly.
For raylib integrations, if application registers a callback with mi_job_percent_register_cb() call, the callback is called during iray/progressive rendering. The zero percentage value is passed if the exact percentage is not known (unlimited rendering), the type parameter of the callback is miJOBP_RENDER. In earlier mental ray versions, this callback was only called for other rendering tasks (tile rendering, photon emission etc.).
In multihosting rendering, reduce shutdown time of slaves/satellites in case of the master host shutdown or crash.
The function mi_tri_vectors() is deprecated (but still functional). Instead, a series of mi_query() modes is provided:
- miQ_PRI_VERTICES_POINTS
- miQ_PRI_VERTICES_NORMALS
- miQ_PRI_VERTICES_MOTION
- miQ_PRI_VERTICES_TEX (accepts space)
- miQ_PRI_VERTICES_USER (accepts space)
- miQ_PRI_VERTICES_BUMPS_U (accepts space)
- miQ_PRI_VERTICES_BUMPS_V (accepts space)
- miQ_PRI_VERTICES_DERIVS_U
- miQ_PRI_VERTICES_DERIVS_V
- miQ_PRI_VERTICES_DERIVS_UU
- miQ_PRI_VERTICES_DERIVS_VV
- miQ_PRI_VERTICES_DERIVS_UV
- miQ_PRI_VERTICES_DERIVS_VU
- miQ_PRI_VERTICES_DERIVS_UV
Caller should provide a space to store results: an array of vectors, one for each vertex of the primitive. In this version, the query modes are functional only for triangle mesh, so it is safe to assume that the number of vectors is 3.
```
     miVector  tri[3];
     miVector *q[] = { &tri[0], &tri[1],  &tri[2]};
     mi_query( <query mode>, state, 0, q, <space for modes accepting space>);
```
Examples of the shader usage are available in the source code of base shaders.
For C++ shader interface, added an utility class mi::shader::State_importance (shader_importance.h header file) as a handy way to adjust importance factor.
During finalgather precomputation, no longer check if the angle between interpolated normal and geometry normal is below 90 degrees and flip geometry normal in such cases. The large angle can be a result of a string bump mapping is applied.
mental ray 3.8.3.23 and earlier versions printed a warning and flipped geometry normal, assuming that the issue originates from an incorrect shader. Many old shaders which computed the irradiance on the back side failed to flip the geometry normal.

The old geometry normal flipping behavior can be enabled by setting the string option "finalgather flip bad geom normal" <bool> to true.
For the .mi parser, added syntax to remove a string option: "optionname" NULL inside of the option block.
Added {_MI_REG_FBVIRTUAL} registry and MI_FBVIRTUAL environment variable for the completeness of -fb_virtual command line option interface to specify if in-memory/memory mapped/cached framebuffers are used.
Added MI_FBDIR environment variable which specifies the directory to keep temporary framebuffer files. (The command line option and {_MI_REG_FBDIR} registry exist in earlier mental ray 3.8 and earlier versions).

Bug fixes:

For BSP2 acceleration, fixed possible crack artifacts on 64-bit platforms. Rare numerical precision issue.
Fixed possible crash in rendering of scenes with assemblies containing shader declarations. If multiple frames were rendered, a crash could occur in a frame following a scene change removing a previously used assembly from the scene DAG.
For rendering of scenes with assemblies containing phenomena declaration inside of assemblies: fixed possible "phenomenon root not found" error messages and black images or image parts in second and subsequent frames.
Fixed possible artifacts and crashes in hair rendering when both rasterizer and ray tracing are enabled. The rasterizer and ray tracing tessellations of hair were mixed up in some cases.
Fixed possible crash in erroneous scenes with phenomena containing undeclared shaders.
For scenes with large number of assemblies, fixed an inefficiency in rasterizer mode.
Fixed a memory leak in loading detailed shadowmaps from file.
In multihosting rendering, fixed a significant memory leak on the slave/satellite hosts in finalgathering precomputation.
The shader C/C++ source code file specified with "code" statement in .mi syntax was removed after rendering. Fixed.
For MetaSL light shaders, fixed sampling of area lights with MetaSL light shaders returning uneven sample weights.
For MetaSL shaders, fixed light normal (wrong coordinate space was used).

Shader changes:

For the production mip_rayswitch_advanced shader, fixed a crash if an attached parameter shader had more than one output. The first output of the latter shader is taken as a result of mip_rayswitch_advanced now.

Changes in version 3.9.0.4

Feature improvements:

Implemented direct parsing of .mi files compressed with gzip. The compression is detected automatically. Added an echo bit and command line option (-echo compress ...) which would echo a scene directly into a gzip compressed file.
Added two new mi_rc_run_query() modes: miRCQ_CUDA_DEVICE_PROPERTIES and miRCQ_CUDA_MEMORY_STATS. The parameter passed to the call should be the two new structures with the matching names added to mirelay.h. On success, mi_rc_run_query fills the structures are with the corresponding values.
Extended .mi syntax for verbatim textures. The texture header can now be given in a form
color texture "texname" [<width> <height> "typename"]

In particular, this allows to specify "rgbe" textures not distinguishable from rgba""rgba" with the old syntax.
Added new mi_rc_run_query() mode miRCQ_IRAY_VERSION. The passed const char ** parameter is initialized with the address of the iray plugin version string if iray plugin is successfully loaded. Note that calling to this function triggers loading of iray plugin.
For mental ray standalone, added -echo command line option qualifier "iray". If used, the scene is echoed after substituting shaders with those supported by the iray rendering mode.
For mental ray standalone, allow http and https URLs to be specified as echo 'file' output parameter.
Improved options info banner printed with progressive rendering. The output format used in standard rendering mode.
For progressive rendering, including iray mode, allow max samples to be set to 0 or a negative value. In this case, the value is treated as "unlimited", and the rendering is stopped by other criteria: error threshold or maximal rendering time. The default value (100) is used if the string option "progressive max samples" is not set.
Build MetaSL LLVM backend on the basis of LLVM 2.6. Earlier version used LLVM 2.5. With this change, a separate LLVM module is created for each shader compiled. In particular, this improves robustness when shaders are deleted or modified.
iray rendering mode is switched to use CUDA 3.0.
For raylib integrations using interactive progressive rendering: added mi_rc_progressive_resume_rendering() function. This allows continuation of the rendering process after mi_rc_progressive_stop_rendering() or mi_rc_progressive_cancel_rendering() calls.
For raylib integrations using interactive progressive rendering: added mi_rc_progressive_region_of_interest() function. When called, upcoming samples will be places inside of the region specified by the function.
Feature and shader coverage set of iray rendering mode has been improved.

Bug fixes:

Fixed a series of issues in MetaSL LLVM backend.
For ccmesh objects, fixed possible wrong subdivision level selected for subobjects.
For mental ray standalone, texture (-T) and include (-I) search paths provided on the command line were combined. Fixed.
For mental ray standalone, fixed command line parsing when multiple -L paths are provided. The path lists are concatenated.
In lightmap baking, lightmap shaders were executed several times if an array of materials with several entries of the same materials was inherited to the lightmap object. Added a test for this case in order to avoid the redundant work.
Fixed performance regression: a light shader return value 2 was ignored and did not terminate the light sampling loops as specified.
Significantly improve performance of image based lighting (IBL) used in combination with finalgathering. The number of environment samples done for finalgather ray hits was unnecessary high.
Fixed artifacts for IBL approximate mode (the mode enabled with "environment lighting mode" string option value "approximate").
In some cases, in particular when BSP2 acceleration was used, available work was not scheduled on all threads, fixed. This improve multi-threading scalability, in particular during the construction phase of the BSP2 acceleration for large scenes.
On Windows platforms, fixed a crash when a shader printed a very long (1024 or more symbols) message with mi_info(), mi_warning(), mi_error() and similar function. The crash was in the shader.lib library. 3rd party shader writers should re-link shader DLLs with the new shader.lib in order to profit from this fix.
For interactive progressive rendering, fixed possible crash in accessing freed camera while changing the framebuffer resolution.
For MetaSL shaders, accept shader graphs with only the surface_bsdf node specified.
Fixed a crash when a MetaSL shaders was used in a frame, and a C/C++ shader library was loaded after that frame was finished.

Shader changes:

For fast subsurface scattering shader, fixed possible artifacts due to missing QMC sampling initialization.
For fast subsurface scattering shader, fixed possible noise an unstable images rendered differently with multiple runs on the same settings. Lightmap textures being rendered are excluded from the usage by reflection / refraction ray hits coming from the same or some other lightmap computations.

Changes in version 3.9.0.3

For mental ray standalone executables and integration libraries protected my SPM, switch the license counting to 1 license per machine. This is more generous than 1 license per 4 threads used since version 3.8.1.16.

Known limitations:

Framebuffer code is undergoing refactoring and may be unstable in this version.
MetaSL compiler is undergoing refactoring and may be unstable in this version.

On MacOSX 10.6 "Snow Leopard", iray does not work in 32-bit mental ray. An application would crash during the plugin loading. 32-bit executables on MacOSX 10.5 are not affected.
(For 64-bit executables, on both MacOSX 10.6 and 10.5, iray will run in CPU-only mode as 64-bit CUDA runtime library is not available).

Changes affecting compatibility:

In shader C/C++ interface, some missing const qualifiers have been added. In particular, the mi_db_access() and mi_db_access_type() functions return const void * pointer now, to hint shader writers that the database elements accessed that way should not be modified. This may require recompilation of some C++ shaders adding a const_cast<>.
miState::cache has been removed from the miState struct. Some volume shader trying to ray trace shadows from the volume may need minor modifications (remove the references to that state member).
iray integration interface has been extended and is not compatible with libraries used with mental ray 3.8.

Feature improvements:

Performance, memory consumption and robustness of the BSP2, acceleration, including motion scenes, has been improved compared to mental ray 3.8 and mental ray 3.9.0.0.
For ccmesh objects, handing of texture seams has been improved. The first texture space takes into account possible texture seams of further texture spaces, and vice versa.
Added functionality to detect the amount of memory available to the application. The estimation is printed out before rendering each frame.
Significantly improved performance of the Map containers.
Following customer requests, the default value for per-framebuffer premultiplied flag introduced in version 3.8.0.8 has been changed to maintain compatibility with mental ray 3.7+ and prior. For the primary framebuffer, the default value is true. For user color framebuffer, the default value is false. The flag is effective for rasterizer mode only.
For IBL, added option to control the IBL brightness. The environment lighting scale" f [f f f] string option affects all illumination from the environment, but not the directly visible environment.
Build MetaSL LLVM backend on the basis of LLVM 2.6. Earlier version used LLVM 2.5. With this change, a separate LLVM module is created for each shader compiled. In particular, this improves robustness when shaders are deleted or modified.
In progressive rendering, including iray rendering mode, mental ray kernel now tracks incremental changes to the camera and immediate (bottom level) camera instance done with mi_api_* calls. This makes the requirement to call the mi_rc_progressive_set_camera() function obsolete.
The list of shaders and features supported by iray has been significantly expanded.
For iray rendering mode, intercept CUDA errors and memory allocation errors. CUDA devices issuing errors are excluded from the rendering.
For iray rendering mode, improved scalability when rendering with multiple GPUs and when both GPU and CPU are used.
For iray rendering mode, improved rendering performance when multiple CUDA devices with different CUDA capabilities are used. Capability 1.1 devices are not longer switched to CUDA capability 1.0 mode in such case.

Bug fixes:

Light loops did not stop on light shaders returning the value 2. Regression, fixed. Performance of IBL in combination with finalgathering has been significantly improved.
For lightmap rendering, avoid running lightmap computation multiple times in case of redundant materials in the material array inherited for the lightmap object.
Fixed possible self-intersection artifacts in the build-in ambient occlusion.
Fixed ray differential initialization in finalgather precomputation stage.
Fixed possible crash at shutdown if API forward declarations were used.
Fixed texture filtering in rasterizer mode. Ray differentials were not initialized properly.
For texture lookups with negative filter value: a single image pixel with no filtering is returned.
Fixed a crash in case of incremental change for a camera stereo mode (on/off).
Fixed possible ccmesh artifacts if non-view dependent approximation was used.
For ccmesh objects, fixed echo of derivatives and length approximations.
For subdivision surfaces (ccmesh implementation), fixed a crash observed for some objects.
Fixed a crash if a hair object with no hairs is rendered.
Fixed a crash if ambient occlusion with caching was used in combination with contour rendering.
Fixed finalgather preview rendering with color profiles. Color profile was not applied.
Fixed an artifact on triangle edges that occurred when using BSP2 acceleration.
Fixed a clipping artifact when using BSP2 acceleration.
Fixed rendering speed regression if BSP2 acceleration used for scenes with builtin IBL.
For the BSP2 acceleration, fixed a memory leak with assemblies and multiple instances.
For scenes with shadow maps, a copy of string option tags was deleted twice. This is not know to cause any observable issues.
Fixed a crash if photon map (globillum or caustic) was used in a scene with assemblies.
Added traversal callback functionality for assemblies.
For mental ray integrations, fixed a race condition which could lead to a crash in aborted photon emission.
Do not apply gamma correction for images of half type, so that half and float images are handled consistently.
Fixed a crash in output of multiple framebuffers into the same OpenEXR file for scenes with more than 6 user framebuffers rendered.
Fixed possible too high capacities (bit per pixel, alpha) used in the main color framebuffer file for scenes with more than 6 user framebuffers rendered. In particular, rgba primary framebuffer could be advanced to rgba_16 without a need to do so, consuming more memory.
In progressive rendering, camera movement was not taken into account when rendered with scanline off. Fixed.
For IBL (image based lighting), fixed possible NaN pixel colors if the "environment lighting shadow" modes "solid" or "off" were used.
When rasterizer was used with very low shutter times (0.0001), the scene was rendered black. Fixed.
For hair ray tracing, state->bary[0] sometimes was initialized to NaN value, fixed.
Calling mi_output_image_open() function with out-of-range framebuffer index could cause an unknown tag access fatal or crash. Fixed, an error message is printed.
For material phenomena with no surface shader, attach bsdf (if any) to the surface slot.
Fixed major memory and database tag leaks for scenes with assemblies. Shader instances defined inside of assemblies were not deleted together with assembly.
Fixed a memory and database tag leak in multihosted rendering. For geometry tessellation jobs executed on slave hosts, the parameter tags were not deleted. Tessellation is likely to be executed on slave hosts for scenes with finalgather.
In network rendering, fixed possible fatal error master host if a connection to a slave/satellite on the master host when a connection to a slave/satellite was closed during the initial handshake. This was observed on Windows platforms if the satellite service port was misconfigured.
In multihosted rendering, fixed possible memory (on Windows platforms - handle) leak on slave hosts.
In multihosted rendering, removed some pointless communication between hosts. This slightly improves performance of the network rendering.
For mental ray standalone, the -core command line option was ignored, and the value was always set to on. Fixed.
For iray, prevent possible crash if iray plug-in was not loaded.
For MetaSL shaders, the value of Light_iterator::dot_nl was not adjusted according to the modification of State::normal (bump mapping). Fixed.
For MetaSL compiler, fixed missing conversion for bool vectors.
For MetaSL compiler, fixed equal/non-equal operators for Ray type.
For MetaSL compiler, fixed expressions in array sizes.
For MetaSL Illumination node shaders, removed deprecated "compile as bsdf" annotation. In particular, this fixes internal shader declaration generated for these shaders.
For MetaSL C++ backend, fixed code generated for 3-dimensional float texture samplers.
For MetaSL shaders, fixed regression where shader state inputs were being ignored.
For MetaSL shader runtime, fixed 4th component of half4/float4/double4 texture sampler lookups.
Fixed a crash when a MetaSL shaders was used in a frame, and a C/C++ shader library was loaded after that frame was finished.
For MetaSL compiler, issue errors if global variables are declared. Such variables are illegal according to the MetaSL spec.
For MetaSL LLVM backend, fixed the search path provided with the {_MI_REG_LIBRARY} register. The register is now correctly taken into account.
For MetaSL LLVM backend, fixed inputs of struct types.
For MetaSL LLVM backend, fixed problem with consecutive compilations.
For MetaSL LLVM backend, evaluate only const and uniform inputs in the constructor. This resolves possible deadlock on the init shader lock.
For MetaSL LLVM backend, fixed possible crash in an input parameter of enum type was used.
For MetaSL LLVM backend, fixed a code generation on 64-bit platforms: call immediate instructions are avoided. Call immediate instructions use 32-bit offsets, which may cause sporadic crashes when LLVM-generated code is located at addresses with are more than 2GB away from the mental ray kernel code.
For MetaSL LLVM backend, fixed runtime crash due to calling generated destructors at improper time.
For MetaSL LLVM backend, fixed wrong code generation for swizzles and for the ternary ( "? :" ) operator.
For MetaSL LLVM backend, fixed bytecode generation for shaders with array of structs parameters.

Shader changes:

For fast subsurface scattering shader, fixed possible artifacts due to missing QMC sampling initialization.
For fast subsurface scattering shader, fixed possible noise / unstable images rendered differently with multiple runs. Lightmap textures in the process of being rendered are excluded from usage by reflection / refraction ray hits coming from the same or some other lightmap computation.
For the architectural material shader, fixed Exact AO feature. Regression from mental ray 3.7+.
Fixed a crash in the lume Glare output shader.
For fast subsurface scattering shader, fixed printing of progress messages handle id's on 64-bit platforms.
Platform specific: for physical subsurface shader, fixed artifacts on 64-bit Windows.
For paint.dll shader library on Windows platforms, copyright/version information was not included. Fixed.

Changes in version 3.9.0.0

Feature improvements:

For the BSP2 acceleration used for motion blur scene, enabled new prototype implementation of a fast motion blur BSP2 mode with a much lower memory consumption compared to previous solutions.

Changes in version 3.8.2.0

Feature improvements:

For progressive rendering, disable clamping of the framebuffer values by default.

Bug fixes:

For SPM-protected mental ray licensed by mental images, fixed the new licensing scheme.

Changes in version 3.8.1.18

Feature improvements:

For iray, finish the separation of iray related functionality out of mental ray into the external libraries libiraymr and libiray. The libraries are loaded automatically by mental ray as soon as iray rendering is enabled, and searched for along the regular library search path.
For iray, enable support for CUDA compute capabilities 1.0 in addition to compute capabilities 1.1. This allows to render with iray even on first generation of CUDA-enabled GPUs, but may take up to 20 percent more time to finish.

Bug fixes:

For raylib integration, expose functionality to convert colorprofiles from the internal colorprofile to a specified colorprofile.
Fixed a crash that occurred with phenomena declarations in assemblies.
Fixed a crash for irradiance particles when increasing the IP indirect passes between frames.
Fixed a crash in imf_disp that occurred when changing the camera resolution or switching stereo mode on incrementally.
Fixed imf_disp with command line option --disable-gl set.

Changes in version 3.8.1.17

Feature improvements:

For iray, shader mapping functionality and component of mental ray connectivity has been moved from mental ray standalone/raylib to a separate dynamic library, libiraymr.so/.dll/.dylib. This library is distributed together with iraylib and is installed at the same location.
For MetaSL compiler, improved compiler error file/line reporting.

Bug fixes:

Fixed another crash for BSP2 acceleration for scenes with motion blur containing both moving and static objects. The fix in version 3.8.1.16 was incomplete.
For the BSP2 acceleration used for motion blur scene, fixed possible improper selection of splitting planes, which could lead to memory consumption inefficiency.
For MetaSL LLVM backend, fixed code generation for enums.
For MetaSL compiler, fixed search path for the import statements.
For MetaSL compiler, fixed texture comparison operators.

Changes in version 3.8.1.16

For SPM-protected mental ray licensed by mental images, simplified the licensing scheme.

One license per 4 cores on a machine is requested.

In particular this new scheme provides a fair license counting on virtual machines, which was not possible to achieve with the per-CPU-socket scheme used before.

Bug fixes:

Fixed a crash for BSP2 acceleration for scenes scenes with motion blur containing both moving and static objects.
When mental ray is low on memory, fixed possible segmentation fault crashes for shaders using mi_tri_vectors() function. The list of such shaders includes architectural material, subsurface scattering shaders, and many common OEM shaders.
The validity scope of mi_tri_vectors() function is adjusted. The returned pointers are valid until the next call to that function for a given rendering tile/task. In particular, shaders may need to be adjusted to copy the results before evaluating shader parameters (mi_eval), tracing rays and calling other functions which would lead to execution of other shaders. In earlier versions of mental ray, the validity scope of the data returned by pointer was not specified and could be extremely short when mental ray was low on memory.
When mental ray is low on memory, fixed possible rare unknown database tag access fatal (rare race condition).
For mi_api_add_shader() function, check if the new tag for the given name is the same as the old one, and issue an error in this case. The usage is illegal as this function is unfenced to delete the old tag.
In debug version of mental ray, suppress messages from the ZLIB library, used in particular in (de)compression of some of OpenEXR images.

Changes in version 3.8.1.15

Bug fixes:

Fixed MetaSL LLVM backend crash on Windows XP machines.
For MetaSL LLVM backend, the default optimization level of LLVM is set to zero. The optimizer in LLVM 2.5 JIT compiler has been observed to produce illegal code in some case.

Changes in version 3.8.1.14

Summarized release notes for iray are provided in a separate document.

Feature improvements:

MetaSL shaders are compiled on demand shortly before first call (LLVM backend or C++ backend with C++ external compiler invocation part of the compilation process). On multi-CPU machines this allows parallel compilation of several shaders.
In earlier versions, compilation was triggered by scene preprocessing and was done sequentially.

Bug fixes:

On MacOSX platform, fixed mental ray crash if LLVM backend is used.
For rendering of hairs in combination with motion blur, fixed possible segmentation crashes, missing axis-aligned hair parts and possible performance drop.
Changed the way shader declarations are re-created.
- If shader declaration is derived from MetaSL shader, the tag is not changed, only its content does.
- Re-declarations of shades inside of assemblies are ignored, this restores mental ray 3.7+ behavior.
Together, this fixes issue with deleting shader declarations for existing shaders.
For Map containers, remove temporary files properly.
In multihosted rendering, removed redundant error message.
Fixed mental ray unknown tag fatal in erroneous situation when a shader declaration (or SL shader) MetaSL shader) has been deleted while some of the shader instances are still present in the database.
Fixed possible crash when MetaSL shaders or phenomena were incrementally changed.
For MetaSL compiler, fixed compilation of type names contain semicolons.
For MetaSL C++ and LLVM backends, pass input parameters to shader constructors and destructors.
For MetaSL C++ backend, fixed compilation of scalar and vector texture samplers, as well as cube texture samplers. Regression in version 3.8.1.8.
For the imf_disp tool, possible wrong image display after window resizing.

Changes in version 3.8.1.13

Known issues in this version:

On MacOSX platform, mental ray will crash if LLVM backend is used (since version 3.8.1.8). Fixed in 3.8.1.14.

Bug fixes:

Removed undocumented and unneeded mi_phen_clone/unclone functions.
For mental ray standalone, fixed a regression in command line parsing which may lead to a crash.
Texture pyramid fix in version 3.8.1.12 was incorrect, a proper fix is added.
For MetaSL shaders, fixed builtin bsdf names according to the MetaSL spec.

Changes in version 3.8.1.12

Bug fixes:

Fixed a crash if a texture with on-demand-created filtering pyramids is loaded twice.
For MetaSL compiler, fixed a regressing in parser. The #define's values wrapped by ()-brackets are parsed correctly now.
For MetaSL compiler, fixed conversion of integer vector parameters.
For MetaSL compiler, allow arbitrary const expressions for array size.
For MetaSL compiler, fixed improper handling of const and uniform attributes for constructor calls.
For mental ray used as a MetaSL standalone compiler (ray -mslc command line option), fixed compilation of BSDF shaders into C++ code.

Changes in version 3.8.1.11

Feature improvements:

Added support for MetaSL material phenomena.
For animation rendering, improved incremental changes on the camera framebuffers. If a new framebuffer is specified for an incremental camera, the list of framebuffers is cleaned. In earlier versions framebuffers were accumulated, which lead to issues when rendering animations of echoed scenes.
For finalgathering, provide a way to enforce finalgather computation of a per object/per instance basis. .mi syntax for objects and instances has been extended with finalgather force <bool>. Usual scene DAG inheritance rules apply. The default value is off.
The miIrrad_options structure has been extended with finalgather_force member , so that a shader has a possibility to overwrite the value from the scene DAG. The default value is 0 (inherit). The value 1 can be used to enable, the value 2 to disable finalgather force computation. This option makes the "state->type = miRAY_LM_VERTEX;" trick obsolete.

If finalgather force is enabled, mental ray will compute finalgather indirect illumination directly at the hit point instead of interpolating it from the finalgather map. This produces more accurate result image at the price of increased rendering time.
With -finalgather_passes 0 command line option, it is now possible to enable the compatibility mode for finalgather map computation used in mental ray 3.6+ and prior (all finalgather points are placed in one pass, somewhat lower memory consumption especially for images with huge resolution).
On Linux platforms, add -m32/-m64 option to the default C++ compiler and linker option list (invoked for MetaSL shader compilation with MetaSL C++ backend enabled). In particular, this simplifies deployment of 32-bit mental ray executable on 64-bit Linux platform. On MacOSX this option has already been used in the earlier mental ray versions.
Improved physical memory utilization by avoiding unnecessary largecontinuous memory blocks. In particular, scenes using finalgather maps and shaders using mi_kdtree_* interface are affected,as well as scenes with BSP2 acceleration in the debug build or in combination with multihosted rendering.
Provide -progressive_max_time F command line option in addition to the equivalent string option. This option limits mental ray progressive rendering and iray execution time.

Bug fixes:

Echo requirements for struct and array of struct shader parameters properly. In particular, fixed missing shader instances and textures passed as values for such parameters.
Fixed echo of per-object min/max sample parameters.
Cylinder area light sources were echoed as discs, fixed.
Fixed echo of subdivision surface vertices with darts.
Fixed -output off command line option: it disabled all image file outputs instead of suppressing executions of output shaders.
For 3-dimensional DDS textures, fixed error message (unsupported image format subtype).
On Windows platforms, fixed "Permission denied" error when creating directory for shader cache.
Fixed a crash in unlinking a shader library if MetaSL LLVM backend has used to create some (other) shaders.
For MetaSL compiler, fixed compilation of BSDFs used as parts of phenomena.
For MetaSL Phenomena provided as xmsl files, set all internal default parameter values properly even when they do not belong to the phenomena external interface.
For MetaSL compiler, fixed code generation for shader classes inherited from base shader classes.
For MetaSL runtime, added missing message passing (trace call returns) for several matrix and vector types.
For MetaSL LLVM backend, fixed selection of shader primary output. In particular, this fixes black image rendering if MetaSL light shaders were used.
For MetaSL C++ backend, fixed compiler crashes for Occlusion_options, Irradiance_options, Light_sampler_options, Light_profiles, Particle_map, and Particle_map_options constructors.
For MetaSL C++ backend, fixed illegal code generated for some overloaded matrix and vector functions.
For MetaSL C++ backend, fixed illegal code generated for the Bsdf type.
For MetaSL C++ backend, fixed illegal code generated for Cube and Spectrum texture samplers.
For MetaSL C++ backend, fixed swizzles for types other than float.
For mental ray used as a standalone MetaSL compiler (ray -mslc command line option), fixed generation of uniform and const constructor attributes.
For iray, fixed possible crash when rendering is canceled.

Changes in version 3.8.1.10

Bug fixes:

Fixed an unknown tag fatal or crash for MetaSL shaders with texture parameters having default values.
For file search path substitutions (like those specified with -T/-texture_path command line option), fixed wrong handling of back slashes used in combination with the '!' sign.
For hair rendering, initialize miState::inv_normal to miFALSE for hair hits. For hairs, the flag is meaningless. However, initializing it improves compatibility with non-hair-specific shaders respecting surface orientation.
Fixed potential crashes at standalone mental ray shutdown and at unloading of raylib.dll. Static destructors are called after mi_raylib_exit(). Some of static destructors could potentially access memory blocks in mental ray memory management systems, which could cause a crash. In order to avoid this type of issues, mental ray memory system is changed to release only internal pages with no used blocks on them. This potentially may cause a memory leak after unloading raylib.dll.
Shader writers and application developers are strongly discouraged from using non-trivial static destructors.

Changes in version 3.8.1.9

Feature improvements:

For mental ray standalone, allow multiple -T / -texture_path commands on the command line.
On MacOSX platforms, mental ray loads shader library with dlopen/dlsym instead of the platform-specific NS* function family.

Bug fixes:

Fixed a performance regression in rasterizer. The regression was due to unnecessary frequent memory allocations / deallocations.
For progressive rendering and iray used in combination with DISP protocol, fixed growing memory consumption during long "single-frame" rendering sessions.
Fixed a crash if mi_query function was called in miQ_PRI_BBOX_MIN/MAX modes with null state->pri. mi_query now returns miFALSE in these cases.
Fixed possible crash if the link statement has been applied on a file of an unknown or unsupported type.
For MetaSL C++ backend, fixed code generation for initialization of inputs of struct types containing texture fields and for inputs that are arrays of structs.
For MetaSL LLVM backend, added missing implementation of uniform and const constructors and destructors.
For MetaSL runtime, fixed orthographic state variable.
Fixed a memory leak in MetaSL compiler.

Changes in version 3.8.1.8

Feature improvements:

On MacOSX platform, added a default substitution rule .so -> .dylib. Old shader libraries with .so extension must be renamed. The consistent .so/.dll/.dylib substitution rules provide .mi scene cross-platform compatibility.
For MetaSL shader cache, add user name to the temporary directory created to keep the cache. This fixes possible permission conflicts if mental ray is used by multiple users on the same machine.

Bug fixes:

Fixed a crash when echoing camera with multiple framebuffers.
Fixed missing echoing of inklink statements.
Avoid echoing of shader declarations derived from MetaSL (msl/xmsl) files.
For BSP2 acceleration, fixed possible infinite loop and artifacts if mi_trace_continue function or volumetric rendering with autovolume mode used.
Implemented missing functionality of MetaSL shader distribution in multihosting rendering. MetaSL shaders are send to slaves in text form and generated by each slave host on demand with its own MetaSL compiler and backend.
For MetaSL compiler, fix boolean control expression in the switch statement.
For MetaSL compiler, fixed improper logic in some of switch/break/continue/return combinations.
For MetaSL C++ backend, generate correct code for constructors of texture samplers used in expressions.
MetaSL compiler errors were printed by a wrong module, fixed.
For mental ray standalone invoked as a metasl compiler in order to (re)generate MetaSL code (-mslc -t msl command line options), fixed missing half/double qualifiers on the literals.
Fixed possible mental ray crash at shutdown if MetaSL compiler was used in the session.
For iray on MacOSX platform, load libiray.dylib (instead of .so) .
On Windows platforms, errors in plugin loading (MetaSL LLVM backend, iray) were silently ignored. Added printing out system error messages.

Changes in version 3.8.1.7

Bug fixes:

Fixed crashes in multihosting rendering (regression in version 3.8.1.6/3.8.1.5). Fixed sending link statements from master to slaves.
On MacOSX platform, load gen_llvm.dylib (not .so) for LLVM MetaSL backend.
For MetaSL C++ backend, fixed significant performance issue for array shader parameters.
For MetaSL C++ backend, fixed texture sample constructors used in expressions. Fixed possible wrong type generated for temporary swizzles.

Changes in version 3.8.1.6

Known issues in this version:

In multihosted rendering, a slave would crash when the second library is send from master to the slave host.
On MacOSX platforms, mental ray searches for LLVM backend under the name gen_llvm.so, not gen_llvm.dylib.

Feature improvements:

Improved sampling of IBL (image based lighting), which results in better quality images and in faster convergence of IBL used in combination with progressive rendering.
On MacOSX platforms, dynamic libraries (shaders, plugins) file extension is switched to .dylib. mental ray still allow shaders to have .so extension on these platforms. On Windows, a default substitution rule .dylib -> .dll is added. On Linux, a default substitution rule .dylib -> .so is added. On MacOSX, .so is not substituted into .dylib in order to provide compatibility with old shaders.
For applications integrating raylib with progressive rendering, changes the mi_rc_progressive_tonemapper_function by adding an opaque pointer argument passed to the callback function registered. Note that this changes binary interface and the applications may require re-compilation as well as minor textual changes if plain C is used.
For iray, provide full support for the mia_physicalsun shader.
For iray, improved overall performance. In particular, the performance of image processing with both DISP and mi_rc_progressive_* interfaces has been improved.
For iray, extended support for several OEM shaders.

Bug fixes:

For mental ray standalone on windows platforms, fixed a memory corruption popup window if rendering is aborted (In particular, by pressing Ctrl-C).
Fixed linking info/progress messages incorrectly printed from the DBN module.
For MetaSL parser, fixed possible incorrect handling of import#import token.
For MetaSL compiler, fixed wrong/missing const expression evaluations.
For MetaSL LLVM backend, fixed memory leaks.
For MetaSL LLVM backend, fixed crashes in half and double texture samplers and matrices.
For MetaSL LLVM backend, fixed handling of default case in switch statements.
For MetaSL LLVM backend, fixed support for light_shadow output of light shaders.

Shader changes:

For base and physics shaders, added missing apply attributes.
For lume shader package, removed the declaration of the LumeTools shader used to GUI attributes only.

Changes in version 3.8.1.4

Feature improvements:

For progressive rendering, lift the restriction that task size had to be a multiple of subsampling size.
For standalone MetaSL compiler option of mental ray (mslc-mslc command line option), added a possibility to output MetaSL files. Unused output of binary MetaSL file (.bmsl) is removed.
For raylib integration, expose progressive rendering API in the plain C interface.
For mental ray standalone, allow to specify the value cpu for the -iray command line option. With -iray cpu option, iray rendering will use CPU mode even if CUDA-capable GPU is present in the system.
iray is switched back to CUDA library version 2.2 in order to improve compatibility with drivers.
For MetaSL LLVM backend, added support for shader parameter initializers.

Bug fixes:

Adjusted module id enumeration to keep same error/warning codes as for mental ray 3.7+ and earlier versions.
For progressive rendering, changed the default value of the "progressive error threshold"string option to 0.05, as documented. The value for earlier versions was 0.1.
For MetaSL compiler, implemented missing compatibility between Spectrum and Color, Spectrum and float3, Spectrum and float4 classes.
For MetaSL compiler, fixed a possible crash if 'const String' type was used.
For MetaSL C++ backend, map MetaSL half literals (like 1.0h) to C++ float literals properly.
For MetaSL LLVM backend, added missing support for invocation as a shadow shader.
For MetaSL LLVM backend, added missing dot(int, int) function, Occlusion_options constructor, and String == and != operators.
For MetaSL LLVM backend, fixed handling of destructors on leaving scopes with break, continue and return operators.
For iray, fixed wrong light transforms.
For iray, fixed superfluous verbosity.

Shader changes:

For precompiled MetaSL shader library, added a set of missing intrinsic shaders.
The paint.mi file no longer links base.so explicitly. Though paint phenomena depend on base shaders, the way to link shader explicitly was inconsistent with rayrc configuration workflow.

Changes in version 3.8.1.3

Feature improvements:

For mental ray standalone used as a MetaSL compiler (-mslc command line option), allow file arguments to be specified after compiler options.
For iray, added support for directional light sources.
The GUI attributes for several shader declaration files are moved into separate files. These GUI files are only used by some of the OEM raylib integrations and are excluded from other release tapes.

Bug fixes:

For raylib integrations, fixed possible unknown tag fatal or crash when framebuffers are deleted.
For Map containers, added removal of temporary swap files in case of rendering abort.
For MetaSL LLVM backend, fixed several issues related to the inlining in the generated code with high optimization level (controlled by the MI_METASL_LLVM_OPTLEVEL environment variable).
The version 3.8.1.2 64-bit Linux platform specific issue with MetaSL LLVM backend has been fixed.
For MetaSL LLVM backend, fixed tangent space code, swizzle constructors for atomic types, and parameters to the sample iterators.
For MetaSL C++ backend, fixed literal string symbol escaping.
For raylib OEM integration, fixed a corruption in the GUI attributes handling.

Changes in version 3.8.1.2

Known platform-specific issue in this version: On 64-bit Linux, the MetaSL LLVM backend asserts for most of MetaSL shaders.

Feature improvements:

On Linux and MacOSX systems, changed the way the external compiler and linker are invoked. The standard /bin/sh shell is used to redirect external tool output (stdout and stderr) into temporary files. The fork-based solution used in previous versions was not reliable due to the extensive usage of phtread library (multiple threads, locks).
The environment variable MI_DEBUG_INVOKE is no longer used.
For MetaSL LLVM backend, environment variable MI_METASL_LLVM_OPTLEVEL could be used to control the optimization. It is recommended to set it to 1 on Windows platforms in order to increase generate code reliability.
For iray, added support for the "progressive error threshold" F string option, which stops rendering when specified quality is reached.
For iray, improved support for light shaders. Currently point, rectangle disc and cylinder light area types are supported. physical_light and mib_light_photometric (including light profiles) light shaders are supported.
For iray, significantly reduced the amount of noise for scenes with IBL (image based lighting).
For iray, improved support of several OEM material shaders.

Bug fixes:

Fixed a possible crash in multihosted rendering for scenes using instance light lists.
Fixed on-demand texture filtering (added in version 3.8.0.11) for textures of tiny resolution.
For the BSP2 acceleration, fixed possible improper handling of visibility flags of assemblies. Regression in mental ray 3.8.
For iray, fixed possible rendering artifacts with more than 4 threads.
For iray, fixed an issue in sampling generation causing significant visual difference between images rendered on GPU and on CPU.
For LLVM backend, fixed texture sampler constructors and texture width/height.

Changes in version 3.8.1.1

Feature improvements:

Extended DISP talk protocol. The frame begin reply string has not the form
frame_begin: <frameno> <xres> <yres>, <is_output_shader>

where is_output_shader is 1 for output shaders and 0 for normal rendering. This allows display applications to distinguish the rendering and postprocessing stages.

Bug fixes:

For the imf_disp tool, fixes creating of the second frame if output shaders are used in the scene.

For MetaSL LLVM backend, fixed type conversion for mental ray scene parameters.

Changes in version 3.8.1.0

Feature improvements:

Added -assert_fatal <bool> command line option. If set to off, mental ray would continue execution after assertion (and possibly crash later). Default is on. This option should be used for debugging purposes only. It may be also used to prevent popup on Windows platforms in batch processing.
For MetaSL shaders, added native support for half type.
For MetaSL shaders, improve mapping of MetaSL types to mental ray types. In particular, added support for matrix types.
Changed signature of the recently added mi_lookup_*_texture_x functions, the filter is passed by value.
For the .mi parser, remove the limit of 4096 characters per line.
For the iray, improved light source handling.
Removed left-overs of the deprecated hardware rendering (mental ray 3.4) from the shader and integration interface.

Bug fixes:

Fixed echo of the premultiplied framebuffer flag.
Allow BSDFs to be nested inside of Phenomena.
Fixed a crash for MetaSL shaders used in combination with multi-hosted rendering.
For the standalone MetaSL compiler functionality, only shaders having brdf_type annotation are compiled as BRDFs even with --brdf command line option is specified.
For MetaSL C++ backend, fixed possible wrong initializers inside of phenomena.
For MetaSL shaders, fixed possible artifacts due to wrong volume in/out stack computation.
For MetaSL shaders, fixed missing support for camera dof_focus.

Changes in version 3.8.0.12

mental ray's ability to flatten shading trees and Phenomena specified with .mi syntax or mental ray API and consisting of nodes which are MetaSL shaders has been disabled for 3.8 release. The value of the registry {_MI_REG_METASL_FLATTEN} is ignored. This does not affect flattening of Phenomena provided as XMSL files.

Feature improvements:

Added support for the combination of stereoscopic rendering in combination with progressive rendering.
For the rasterizer, changed the default value of the premultiplied framebuffer flag added in version 3.8.0.8 to true. The new default behavior is backward-compatible with mental ray 3.7 and rendering with rasterizer disabled.
Improved performance for progressive rendering and iray used with DISP callbacks and protocols.
For the MetaSL LLVM backend, added support for the explicit shader calls from MetaSL shaders.

Bug fixes:

For iray, fixed possible crash for scenes using textures.

Changes in version 3.8.0.11

Feature improvements:

iray functionality has been encapsulated into a plugin. This removes the symbol dependencies on the CUDA runtime from OEM raylib libraries. The plugin is name is "iraylib.dll/.so". mental ray searches for this library on the path list for the shader lookup (same mechanism as for the gen_llvm MetaSL backend). The plugin is loaded automatically.
Standalone and SPM-protected mental ray libraries are now delivered with iray plugin and CUDA runtime libraries. The plugin installation is required only for rendering in the iray mode ( -iray command line option).
On 64-bit MacOSX platform, provide CPU fallback functionality for iray.
Provide MIPMAP pyramids for effective filtered texture lookup on demand. Like in the earlier version, mental ray will create MIPMAP pyramids for textures declared filtered at the load/memory map/cache insertion time. If a texture is not declared filtered, the pyramids are created on demand when mi_lookup_color_texture(_x) or mi_lookup_filter_color_texture(_x) function are called for these textures and the upper pyramid levels would improve the performance. Same applies for MetaSL shader filtered texture lookups.
The original image data is not changed, so on-demand MIPMAP pyramids are not available for shaders using low-level mi_img_* interface.
For MetaSL compiler and C++ backend, added support for user data.
For MetaSL LLVM backend, added support for a plurality of features, including mental ray - MetaSL parameter conversions and compilation of light shaders.

Bug fixes:

Fixed a regression: shader cache did not update changed metasl code.
Fixed a regression: in network rendering, the link statements for compiled C/C++ (.so/.dll files) shaders (.so/.dll files) were not propagated from the master host to the slaves.
For MetaSL integration, fixed possible wrong default parameter values in the XMSL phenomena emulation mode (registry {_MI_REG_METASL_EMULATE_MI_PHEN} set on).
Fixed possible false error message on camera stereo mode.
The function void mi_msg_usleep(int secs); is restored in the integration interface.

Shader changes:

Improved architectural shader applied to the glass and similar materials. If after several bounces inside of a glass the trace depth limit has been reached, the black or environment color has been returned. In bright environments this lead to visible "light leak" artifacts. The shader behavior is changed to trace a transparent ray instead of picking environment or black color. It is a better approximation to the true result of several additional reflection/refraction light bounces. For compatibility reasons, the old behavior still could be achieved by setting the per-shader trace depth limit to the value "-1".

Changes in version 3.8.0.10

Feature improvements:

For the BSP2 acceleration and the special case of the shutter interval [0,0] with motion blur option enabled, use static acceleration structure in order to improve performance and reduce memory consumption.
For the MetaSL C++ backend, added support for displacement shaders.
On Windows platforms, removed run-time overhead related to the discontinued MetaSL CIL backend. mental ray kernel no longer relies on the .Net installation.

Bug fixes:

On Windows platforms, fixed a crash when MetaSL compiler LLVM backend was initialized. The functionality of LLVM is still limited, only simple shaders can be compiled in this version.
Fixed possible deadlock or assertion in the debug mental ray version. The issue has been observed if a MetaSL shader has been deleted, or if the IBL feature was used.
Expose mi_rc_run_query function to the shader interface. Though this function is intended for the integration, some existing mental ray shaders depend on it.
For IBL rendering, fixed switching IBL modes between frames.
Fixed a crash in multihosted rendering if finalgather was used.
For MetaSL shaders, fixed possible crash in filtered texture lookup.
For MetaSL shaders compiled with C++ backend, fixed possible improper handling of vector parameters.

Changes in version 3.8.0.9

Feature improvements:

For iray integrated for into OEM libraries, the fallback into the CPU solution has been implemented for Windows and 32-bit MacOSX in addition to Linux.
For MetaSL shaders and C++ backend, added support for volume shaders.

Bug fixes:

Fixed a crash if an attempt to compile MetaSL shader with non-existing compiler backend was done.
Fixed possible crash if mental ray 3.5 compatibility framebuffer syntax was used (regression in version 3.8.0.8).
For imf_disp tool, removed possible display flickering.

Changes in version 3.8.0.8

New features:

MetaSL LLVM backend. A new backend for MetaSL shaders is provided. This backend replaces CIL backend on Windows platforms (superior performance and uniformity of the solution on all platforms). The backend is supported on Windows, Linux and MacOSX: 32 and 64 bit each.
The support for MetaSL CIL backend on Windows is discontinued.

As the name implies, the backend is based on the LLVM technology. There is no dependency on external compilers or linkers. The compilation is done on demand in memory, no intermediate files are created. Currently, it is not possible to externalize shaders compiled with the LLVM backend. Thus, such shaders are not stored into the compiled shader cache. However, in case a shader has been compiled with C++ backend (using external C++ compiler), the existing shaders in the compiled shader cache would be used. Shaders compiled with LLVM and with C/C++ compiler can be mixed in one scene, phenomenon or shading tree.

In order to enable MetaSL LLVM backend, it is necessary to:
- Set the registry " {_MI_REG_METASL_BACKEND}" value to "LLVM" or "llvm".
- Make sure that the file gen_llvm.so / .dll is present in the path list where mental ray searches for shaders. This paths are specified by the -L command line option and by the "{_MI_REG_LIBRARY}" registry value.
LLVM MetaSL backend library is loaded automatically when the first MetaSL shader is compiled.

This version of LLVM backend is an initial implementation, only simplest shaders will compile successfully Known issue: on windows platforms, gen_llvm will crash at initialization.
OEM integration libraries now contain a functional version of iray on 32 and 64 bit Windows and Linux, as well as on 32-bit MacOSX. The 64-bit version of MacOSX is planned.
In this release, iray is linked statically into the mental ray library. T

On Linux system, an automatic detection of the presence of CUDA-capable graphics card is provided. If not available, CPU mode is enabled. This functionality is planned for Windows and MacOSX as well.

In this version, iray relies on CUDA compute capability version 1.1.

Feature improvements:

Added new function
```
    miBoolean mi_trace_probe_x(
        miState        state,
        const miVector *direction,
        const miVector *origin,
        miUint         flag );
```
This function is similar to mi_trace_probe, but takes a bitmask of visibility flags to test hits for. The enumeration for the flags is added to the shader.h interface file.
For the rasterizer, allow per-framebuffer control over the compositing. The new per-framebuffer modifier "premultiplied" on/off has been added to the .mi syntax. If set, the compositing of a color framebuffer(primary or user) is done according to the formula (A + (1-opacity) * B), otherwise it is (opacity * A + (1-opacity) * B).
For MetaSL shaders, added support for ddx/ddy. The expressions only apply for normals, points and texture coordinates. Note that the normal derivatives are computed for the interpolated normal without the effect of bump shaders.
For the imf_disp tool, added "always on top" functionality.
For the imf_disp tool, added mouse wheel scrolling of the window.
For the imf_disp tool, and DISP talk protocol, added the ability to specify region of interest in the progressive rendering mode.
For the imf_disp tool, display currently rendered rectangle.
For the imf_disp tool used for stereoscopic images, allow to select one of the possible ways to display stereo images (View menu item):
- Color Anaglyph (default)
- Grey Anaglyph
- Half Color Anaglyph
- Left Only
- Right Only
For Irradiance Particles, allow to specify number of secondary rays with additional parameters to the -ip_rays command line option. Before it was only possible with a .mi file string option.
Missing const qualifiers are added to the public API.
Improved runtime performance for MetaSL shaders.

Bug fixes:

Fixed possible wrong illumination if IBL was used in a combination with other light sources.
Fixed possible artifacts in motion blur scenes rendered with BSP2 acceleration.
Fixed missing visible lights in scenes with no geometry.
For applications integrating raylib, fixed a possible deadlock if a GUI element with the same name is re-defined without deleting the old one.
Fixed possible crash in rendering of multiple frames with incremental camera changes which switch stereoscopic rendering on/off.
Put some memory blocks allocated with global new/new[] operators under mental ray memory management/statistics system.
For integrated version of mental ray, fix the table of exported symbols needed for the integration.
For the imf_disp tool, link EXPAT library statically. The dynamic library is missing on some older Linux system.
For the imf_disp tool, fixed several image displaying issues.
For MetaSL shaders, fixed a series of bugs related to type conversions and multiple output.
For MetaSL shaders, fixed inefficiently in the coordinate system transformations.

Changes in version 3.8.0.7

Known limitations in this version:

Type system and parts of the MetaSL compiler SDK and runtime system (including headers) have been refactored. Work in progress. In particular, CIL backend currently fails to compile many shaders.
Performance of MetaSL shaders using array parameters is very poor.

Feature improvements:

Added a support for number of area light samples, which is not an exact power of 2. Prior to this version, mental ray would increase the number of sample for area lights. In the current implementation, the exact number of samples to be taken is respected up to the u * v = 64. In order to keep old behavior, the number of samples per light in the scene could be adjusted to be an exact power of 2.
Added echo for the .msl / .xmsl file include statements.
Missing const qualifiers are added to the public API.

Bug fixes:

For IBL, fixed possible shadow artifacts. The shadow rays are shot from the lights towards telemetry, like this is done for usual lights. This increases compatibility of IBL with existing shadow shaders.
On 64-bit Linux machines, fixed possible memory corruption crash. The crash has been in particular been observed during rendering of large scenes with scanline on.
Fixed possible crash in raylib integration. If an application was low on memory and a new thread has been registered with raylib, the application memory handle callback could be called in order to inform application that some memory should be freed if possible. At the moment of the call, the thread is not registered into raylib yet and thus no calls to mental ray functions should be done. In order to prevent possible crash, added a check to mi_info, mi_warning, mi_error etc.: if called from threads not attached to mental ray, these functions will return immediately without printing any messages.
For the SPM-protected mental ray executables and libraries, fixed possible failure to obtain the necessary number of licenses in core-restricted environments (such as virtual machines) and some machines with Intel i7 CPUs.
On Windows platforms, one of the modules has been compiled with platform STL headers, which caused symbol conflicts for an integrated version of mental ray library. Fixed. The complete mental ray now is compiled with STLPORT put into the namespace MISTD. In particular, it does not have a dependency on the _SECURE_SCL define value.
Fixed possible creases during parallel compilation of several MetaSL shaders.

Changes in version 3.8.0.12

Feature improvements:

Added support for the combination of stereoscopic rendering in combination with progressive rendering.
For the rasterizer, changed the default value of the premultiplied framebuffer flag added in version 3.8.0.8 to true. The new default behavior is backward-compatible with mental ray 3.7 and rendering with rasterizer disabled.
Improved performance for progressive rendering and iray used with DISP callbacks and protocols.
For the MetaSL LLVM backend, added support for the explicit shader calls from MetaSL shaders.

Bug fixes:

For iray, fixed possible crash for scenes using textures.

Changes in version 3.8.0.11

Feature improvements:

iray functionality has been encapsulated into a plugin. This removes the symbol dependencies on the CUDA runtime from OEM raylib libraries. The plugin is name is "iraylib.dll/.so". mental ray searches for this library on the path list for the shader lookup (same mechanism as for the gen_llvm MetaSL backend). The plugin is loaded automatically.
Standalone and SPM-protected mental ray libraries are now delivered with iray plugin and CUDA runtime libraries. The plugin installation is required only for rendering in the iray mode ( -iray command line option).
On 64-bit MacOSX platform, provide CPU fallback functionality for iray.
Provide MIPMAP pyramids for effective filtered texture lookup on demand. Like in the earlier version, mental ray will create MIPMAP pyramids for textures declared filtered at the load/memory map/cache insertion time. If a texture is not declared filtered, the pyramids are created on demand when mi_lookup_color_texture(_x) or mi_lookup_filter_color_texture(_x) function are called for these textures and the upper pyramid levels would improve the performance. Same applies for MetaSL shader filtered texture lookups.
The original image data is not changed, so on-demand MIPMAP pyramids are not available for shaders using low-level mi_img_* interface.
For MetaSL compiler and C++ backend, added support for user data.
For MetaSL LLVM backend, added support for a plurality of features, including mental ray - MetaSL parameter conversions and compilation of light shaders.

Bug fixes:

Fixed a regression: shader cache did not update changed metasl code.
Fixed a regression: in network rendering, the link statements for compiled C/C++ (.so/.dll files) shaders (.so/.dll files) were not propagated from the master host to the slaves.
For MetaSL integration, fixed possible wrong default parameter values in the XMSL phenomena emulation mode (registry {_MI_REG_METASL_EMULATE_MI_PHEN} set on).
Fixed possible false error message on camera stereo mode.
The function void mi_msg_usleep(int secs); is restored in the integration interface.

Shader changes:

Improved architectural shader applied to the glass and similar materials. If after several bounces inside of a glass the trace depth limit has been reached, the black or environment color has been returned. In bright environments this lead to visible "light leak" artifacts. The shader behavior is changed to trace a transparent ray instead of picking environment or black color. It is a better approximation to the true result of several additional reflection/refraction light bounces. For compatibility reasons, the old behavior still could be achieved by setting the per-shader trace depth limit to the value "-1".

Changes in version 3.8.0.10

Feature improvements:

For the BSP2 acceleration and the special case of the shutter interval [0,0] with motion blur option enabled, use static acceleration structure in order to improve performance and reduce memory consumption.
For the MetaSL C++ backend, added support for displacement shaders.
On Windows platforms, removed run-time overhead related to the discontinued MetaSL CIL backend. mental ray kernel no longer relies on the .Net installation.

Bug fixes:

On Windows platforms, fixed a crash when MetaSL compiler LLVM backend was initialized. The functionality of LLVM is still limited, only simple shaders can be compiled in this version.
Fixed possible deadlock or assertion in the debug mental ray version. The issue has been observed if a MetaSL shader has been deleted, or if the IBL feature was used.
Expose mi_rc_run_query function to the shader interface. Though this function is intended for the integration, some existing mental ray shaders depend on it.
For IBL rendering, fixed switching IBL modes between frames.
Fixed a crash in multihosted rendering if finalgather was used.
For MetaSL shaders, fixed possible crash in filtered texture lookup.
For MetaSL shaders compiled with C++ backend, fixed possible improper handling of vector parameters.

Changes in version 3.8.0.9

Feature improvements:

For iray integrated for into OEM libraries, the fallback into the CPU solution has been implemented for Windows and 32-bit Macosx in addition to Linux.
For MetaSL shaders and C++ backend, added support for volume shaders.

Bug fixes:

Fixed a crash if an attempt to compile MetaSL shader with non-existing compiler backend was done.
Fixed possible crash if mental ray 3.5 compatibility framebuffer syntax was used (regression in version 3.8.0.8).
For imf_disp tool, removed possible display flickering.

Changes in version 3.8.0.8

New features:

MetaSL LLVM backend. A new backend for MetaSL shaders is provided. This backend replaces CIL backend on Windows platforms (superior performance and uniformity of the solution on all platforms). The backend is supported on Windows, Linux and MacosX: 32 and 64 bit each.
The support for MetaSL CIL backend on Windows is discontinued.

As the name implies, the backend is based on the LLVM technology. There is no dependency on external compilers or linkers. The compilation is done on demand in memory, no intermediate files are created. Currently, it is not possible to externalize shaders compiled with the LLVM backend. Thus, such shaders are not stored into the compiled shader cache. However, in case a shader has been compiled with C++ backend (using external C++ compiler), the existing shaders in the compiled shader cache would be used. Shaders compiled with LLVM and with C/C++ compiler can be mixed in one scene, phenomenon or shading tree.

In order to enable MetaSL LLVM backend, it is necessary to:
- Set the registry " {_MI_REG_METASL_BACKEND}" value to "LLVM" or "llvm".
- Make sure that the file gen_llvm.so / .dll is present in the path list where mental ray searches for shaders. This paths are specified by the -L command line option and by the "{_MI_REG_LIBRARY}" registry value.
LLVM MetaSL backend library is loaded automatically when the first MetaSL shader is compiled.

This version of LLVM backend is an initial implementation, only simplest shaders will compile successfully Known issue: on windows platforms, gen_llvm will crash at initialization.
OEM integration libraries now contain a functional version of iray on 32 and 64 bit Windows and Linux, as well as on 32-bit Macosx. The 64-bit version of Macosx is planned.
In this release, iray is linked statically into the mental ray library. T

On Linux system, an automatic detection of the presence of CUDA-capable graphics card is provided. If not available, CPU mode is enabled. This functionality is planned for Windows and Macosx as well.

In this version, iray relies on CUDA compute capability version 1.1.

Feature improvements:

Added new function
```
    miBoolean mi_trace_probe_x(
        miState        state,
        const miVector *direction,
        const miVector *origin,
        miUint         flag );
```
This function is similar to mi_trace_probe, but takes a bitmask of visibility flags to test hits for. The enumeration for the flags is added to the shader.h interface file.
For the rasterizer, allow per-framebuffer control over the compositing. The new per-framebuffer modifier "premultiplied" on/off has been added to the .mi syntax. If set, the compositing of a color framebuffer(primary or user) is done according to the formula (A + (1-opacity) * B), otherwise it is (opacity * A + (1-opacity) * B).
For MetaSL shaders, added support for ddx/ddy. The expressions only apply for normals, points and texture coordinates. Note that the normal derivatives are computed for the interpolated normal without the effect of bump shaders.
For the imf_disp tool, added "always on top" functionality.
For the imf_disp tool, added mouse wheel scrolling of the window.
For the imf_disp tool, and DISP talk protocol, added the ability to specify region of interest in the progressive rendering mode.
For the imf_disp tool, display currently rendered rectangle.
For the imf_disp tool used for stereoscopic images, allow to select one of the possible ways to display stereo images (View menu item):
- Color Anaglyph (default)
- Grey Anaglyph
- Half Color Anaglyph
- Left Only
- Right Only
For Irradiance Particles, allow to specify number of secondary rays with additional parameters to the -ip_rays command line option. Before it was only possible with a .mi file string option.
Missing const qualifiers are added to the public API.
Improved runtime performance for MetaSL shaders.

Bug fixes:

Fixed possible wrong illumination if IBL was used in a combination with other light sources.
Fixed possible artifacts in motion blur scenes rendered with BSP2 acceleration.
Fixed missing visible lights in scenes with no geometry.
For applications integrating raylib, fixed a possible deadlock if a GUI element with the same name is re-defined without deleting the old one.
Fixed possible crash in rendering of multiple frames with incremental camera changes which switch stereoscopic rendering on/off.
Put some memory blocks allocated with global new/new[] operators under mental ray memory management/statistics system.
For integrated version of mental ray, fix the table of exported symbols needed for the integration.
For the imf_disp tool, link EXPAT library statically. The dynamic library is missing on some older Linux system.
For the imf_disp tool, fixed several image displaying issues.
For MetaSL shaders, fixed a series of bugs related to type conversions and multiple output.
For MetaSL shaders, fixed inefficiently in the coordinate system transformations.

Changes in version 3.8.0.7

Known limitations in this version:

Type system and parts of the MetaSL compiler SDK and runtime system (including headers) have been refactored. Work in progress. In particular, CIL backend currently fails to compile many shaders.
Performance of MetaSL shaders using array parameters is very poor.

Feature improvements:

Added a support for number of area light samples, which is not an exact power of 2. Prior to this version, mental ray would increase the number of sample for area lights. In the current implementation, the exact number of samples to be taken is respected up to the u * v = 64. In order to keep old behavior, the number of samples per light in the scene could be adjusted to be an exact power of 2.
Added echo for the .msl / .xmsl file include statements.
Missing const qualifiers are added to the public API.

Bug fixes:

For IBL, fixed possible shadow artifacts. The shadow rays are shot from the lights towards telemetry, like this is done for usual lights. This increases compatibility of IBL with existing shadow shaders.
On 64-bit Linux machines, fixed possible memory corruption crash. The crash has been in particular been observed during rendering of large scenes with scanline on.
Fixed possible crash in raylib integration. If an application was low on memory and a new thread has been registered with raylib, the application memory handle callback could be called in order to inform application that some memory should be freed if possible. At the moment of the call, the thread is not registered into raylib yet and thus no calls to mental ray functions should be done. In order to prevent possible crash, added a check to mi_info, mi_warning, mi_error etc.: if called from threads not attached to mental ray, these functions will return immediately without printing any messages.
For the SPM-protected mental ray executables and libraries, fixed possible failure to obtain the necessary number of licenses in core-restricted environments (such as virtual machines) and some machines with Intel i7 CPUs.
On Windows platforms, one of the modules has been compiled with platform STL headers, which caused symbol conflicts for an integrated version of mental ray library. Fixed. The complete mental ray now is compiled with STLPORT put into the namespace MISTD. In particular, it does not have a dependency on the _SECURE_SCL define value.
Fixed possible creases during parallel compilation of several MetaSL shaders.

Changes in version 3.8.0.6

Feature improvements:

For the integrations of raylib progressive rendering: the mi_rc_progressive_set_buffer() function get an additional bool argument with the default value false. If the new parameter is set to true, multiple calls to mi_rc_run() could be used with the same framebuffer,allowing a simplified and more efficient call sequence as in the pseudo-code below. The call to mi_rc_progressive_shutdown() is not required in this case.
```
    mi_rc_run(miRENDER_PREPROC|miRENDER_RENDER| ..., ... );
    mi_rc_progressive_set_buffer( ... settings ... , true);
    while (loop condition) {
        mi_rc_progressive_cancel_rendering() or mi_rc_progressive_stop_rendering();
        // do application incremental updates 
        // BSP is *not* rebuild by the kernel
        mi_rc_run(miRENDER_REINHERIT | 
                  [miRENDER_REINHERIT_LIGHTS |]
                  miRENDER_RENDER, ...);
    }
    mi_rc_run(miRENDER_POSTPROC, ...);
```
As the binary interface of the mi_rc_progressive_set_buffer() function has been changed, applications with progressive rendering integration need to be recompiled against the new header files. No source code changes are required for application using the old call sequence.
Reduced memory occupied by BSP2 acceleration for motion blur scenes.
The files generated for the debugging purposes if the registry "{_MI_REG_METASL_TRACE}" is set are put into the temporary files directory instead of current directory.

Bug fixes:

Fixed possible crash in raylib integration. If an application was low on memory and a new thread has been registered with raylib, the application memory handle callback could be called in order to inform application that some memory should be freed if possible. At the moment of the call, the thread is not registered into raylib yet and thus no calls to mental ray functions should be done. In order to prevent possible crash, added a check to mi_info, mi_warning, mi_error etc.: if called from threads not attached to mental ray, these functions will return immediately without printing any messages.
If a shader declaration was deleted, the DLL library containing the code for the corresponding shader implementation has been unlinked. Limit unlinking of the DLLs to the case of MetaSL shaders compiled by mental ray.
Put some memory blocks allocated with global new/new[] operators under mental ray memory management/statistics system.
Fixed possible crash in erroneous scenes with undefined material inherited as a component of material array in a scene DAG.

Changes in version 3.8.0.4

Bug fixes:

For the rasterizer, fixed possible artifacts for unsplitted quads crossing the clipping plane. The fix is complimentary to the fix in version 3.8.0.3.

Changes in version 3.8.0.3

Feature improvements:

Experimental. Added an experimental extension to the DISP talk protocol and imf_disp tool.
The command refine_fb xl,yl,xh,yh can be used to specify that in the progressive rendering mode, additional samples are placed in the specified window if progressive rendering is used. imf_disp tool allows the selection of the window with the mouse.
Added unlink "filename" command to the .mi syntax in order to allow shader library replacements in mental ray standalone. Without this syntax, this functionality was available for raylib integration only.
For scenes with motion blur, improved BSP2 construction time for objects having single motion segment.
For MetaSL C++ backend, improves support for constructors and destructors.

Bug fixes:

For integrations of raylib containing iray, fixed possible symbol conflicts.
For integration of raylib in a form of static library, fixed possible missing debug information due to several pairs of code object files with the same names. Relevant for Windows platforms only.
For the progressive rendering displayed with the imf_disp tool, fixed stopping of the rendering. During the progressive rendering, the "stop" (red square) button of the playback button group can be used to signal mental ray to stop the rendering.
For ccmesh objects, fixed memory allocations bypassing mental ray memory manager. This memory was not counted in statistics and could lead to a crash for large scenes.
Fixed possible ray hit misses for BSP2 acceleration used for photons (globillum and caustics).
Fixed a regression: registry substitution has not been applied when the type of linked library files has been determined, causing disfunction of .mi statements like link "base.{EXT}".
For the rasterizer, fixed possible artifacts for quads crossing clipping plane.

Changes in version 3.8.0.2

Feature improvements:

For MetaSL shaders, implemented IOR in/out stack functionality.

Bug fixes:

Restored stop command and stop button functionality in imf_disp 2.0 for the progressive rendering mode.
Fixed missing indirect illumination of finalgathering was used in combination with importons for scenes containing materials using BSDFs.
Fixed a crash if finalgathering was used in combination with importons, and the importon map was empty.
For MetaSL shaders used with C++ backend, fixed passing parameters over the trace calls.
For MetaSL shaders used with C++ backend, generate correct code for the conversion of shader type parameters to boolean.

Changes in version 3.8.0.1 (incomplete list).

Extended list and more detailed description for new features and changes will be provided with the next version.

Iray:

Some mental ray 3.8 distributions are coming with iray. Iray is a CUDA-based high performance path tracing renderer implementing mental images architectural material and material library shading model. The integration of iray into mental ray follows the raylib progressive rendering interface. In order to enable iray, the command line -iray on options, or a string option "iray" on should be used. iray is work in progress and is not functional in this alpha version.

New features:

Stereoscopic rendering. mental ray 3.8 provides a workflow to render two images (for the left and for the right eyes) simultaneously. Two rendered images are stored in two separate framebuffers sets and saved as pair(s) of image files. The names of the files are prefixed with lft_ and rgt_ for the left and right eye respectively. Stereoscopic rendering is supported for the rasterizer as well as scanline on/off. View-dependent tessellation and displacement is applied according to the camera located in the middle of two eye positions. Same applies for the computation of finalgather maps. This allows consistent rendering of two images and improves performance by exploiting locality of the data and avoiding redundant work.
The .mi syntax for the camera is extended with

camera "mycamera"

stereo <method> <eye_separation>

...

end camera

where method of left/right eye camera could be one of the following:
- off Default, no stereoscopic rendering.
- toein The camera frustums are rotated so that the two camera direction vectors meet at the focal length. This method introduces perceptually incorrect vertical parallax which increases out from the center of the projection plane and might cause a visual discomfort when looking at rendered stereo images for a longer time.
- offaxis This method is the perceptually correct way to create stereo camera pairs. It introduces no vertical parallax and is therefore less stressful to the observer. It requires an asymmetric camera frustum, which most manufactured cameras do not support as of today.
- offset This is the straightforward method for rendering stereo pairs. It shifts the eyes along the horizontal axis in camera space.
eye_separation is the distance between two eyes in camera space.
It is possible to specify stereoscopic rendering parameters wit the command line option

-stereo <method> <eye_separation>
The integration/shader API has been extended with miCamera::miCamera_stereo (enumeration) and miCamera::eye_separation members.
Stereoscopic rendering is not supported for orthographic cameras. For shaders, the camera space used if the position/orientation of the camera in the middle of two eye.

New imf_disp 2.0 image viewer has been written. Unlike version 1, the implementation is identical on Windows, Linux and MacosX platform, delivering the same set of controls.

The major improvements include the ability to display live any of the rendering framebuffers (not just the primary color one), zoom, display individual channels, expose and gamma control. Automatic tonemapping is provided.

The new image viewer tool is compatible with progressive rendering and with stereoscopic images.

The compete list of command line options is given below:

imf_disp [/l <str>] [/c <str>] [/e <str>] [/g <str>] [/t] [/s] 
            [--frame <num>] [--frame-rate <num>] [--live-update <str>] 
            [--cache-all-layers <str>] [--view-controls <str>] [--playback-controls <str>] 
            [--on-image-load <str>] [--disable-gl] [/h] [image files...]

   -l, --layer=<layername>              layer to display
   -c, --channel=<c/r/g/b/a>            channel to display
   -e, --exposure=<exp>           set exposure
   -g, --gamma=<gamma>              set gamma
   -t, --tone-map            enable tone mapping
   -s, --stereo-display      enable stereo display
   --frame=<frame_number>                  frame to display
   --frame-rate=<fps>             set frame rate (FPS)
   --live-update             monitor file for changes
   --cache-all-layers=<on/off>       retrieve and cache all layers
   --view-controls=<on/off>          toggle displaying the view toolbar
   --playback-controls=<on/off>      toggle displaying playback toolbar
   --on-image-load=<gui_behavior>          set GUI behavior on image load
   --disable-gl              disable GL-based display

Feature improvements:

BSP2 ray tracing acceleration is made the default one.
For scenes without motion blur, significant to performance improvement and dramatically lower memory consumption have been achieved.

For scenes with motion blur, moderate improvements on performance and memory consumption have been achieved. Further improvements are planned before mental ray 3.8 final version.
Hair rendering improvements.
For hair rendering used with ray tracing, a new splitting scheme is used. The scheme resolves possible rendering artifacts known from earlier mental ray versions and moderately improves the performance.

No longer required MI_HAIR_SPLITTING environment variable is removed.

Rendering performance of scenes with many hair objects has been improved.

For hair rendering used in combination with the rasterizer, the new string option "rast hair disposable" <bool>has been added. Default is off. If enabled, mental ray will tessellate only hair parts inside of the current rendered tile view frustum and throw them one one the tile is rendered. This mode is recommended for scenes with many hair as the memory consumption is decreased significantly at a price of slightly longer rendering time.
Progressive rendering, in particular the progressive rendering API, has been improved. For convenience, the summary of the options and API is given below. The complete list of options of relevant string options is given below:
- "progressive" <bool> Turn progressive rendering on or off. The default is off.
- "progressive subsampling size" <int> Enables and controls coarse sampling of the initial displayed images. The default is 0 for disabled. Values greater than 1 activate subsampling in blocks of pixels. The value specifies the size of pixel blocks (size x size) where initial samples are placed first, typically displaying as a coarse image with the impression of a lower resolution. The rendered tile size is adjusted to be a multiple of this subsampling size.
- "progressive min samples" <int> Set the minimum number of samples per pixel to compute before considering any of the listed criteria to stop progressive rendering. The default is 4.
- "progressive max samples" <int> Set number of samples per pixel to stop progressive rendering. The default is 100.
- "progressive max time" <int> Set time, in seconds, when to stop progressive rendering. The default is to run unlimited. Rendering is not guaranteed to stop exactly at the time specified. Especially for the low limits, certain minimal time exceeded those limits may be required.
- "progressive error threshold" <float> Set error threshold to stop progressive rendering. The default is 0.05. A value of 0.0 would apply no quality-driven rendering termination. A value of 0.5 will stop rendering already at a very low quality.
- "progressive interactive" <bool> For the integrated version of raylib, enables a high performance image feedback. Rendering can be displayed and modified interactively via the mi_rc_progressive_* set of integration functions. The DISP image mechanism and disp protocols are disabled for the performance reasons. Default is off.
For mental ray library, the following integration API is provided:
- Set the display buffer and start rendering.
```
extern "C" bool mi_rc_progressive_set_buffer(
    void *display_buffer,
    const size_tstride,
    const miRc_prog_buffer_typetype,
    const boolflip);
 
```
  This function need to be called by the application after calling mi_rc_run(). It provides the buffer used by the application for display to the core and implicitly starts rendering.
  Two types of buffers are supported, 8 bit and floating point.
```
typedef enum {
    PROG_BUFFER_RGBA_8 = 0,// RGBA, 8bit
    PROG_BUFFER_RGBA_32= 1,// RGBA, 32bit fp
} miRc_prog_buffer_type;
```
  Remark: different pixel layouts (like BGRA) could be implemented by providing a tonemapper function that swaps channels.
- Update the display. The application can ask mental ray core at any time if more samples have been computed to be displayed. The recommended workflow is to query the number of computed frames via mi_rc_progressive_get_update_count() call with certain time intervals to see if there is something new to display. If a new frame is computed, mi_rc_progressive_update_buffer() can be used to copy its data to the application's display buffer. The special method mi_progressive_update_buffer_first_subframe() will always return the first subsampled frame, which allows to display constant image while moving the camera (which can be less visually disturbing to the user).
```
extern "C" miUint mi_rc_progressive_get_update_count();
extern "C" miUint mi_rc_progressive_update_buffer();
extern "C" miUint mi_rc_progressive_update_buffer_first_subframe();
 
```
  There are special return values to the calls to above:
```
enum {
    // The rendering is not running (not yet started or already 
    // been shut down)
    MI_RC_PROGRESSIVE_NOT_RUNNING = 0xfffffff,
    // The rendering has finished rendering and is on hold until the
    // application calls mi_rc_progressive_shutdown()
    MI_RC_PROGRESSIVE_FINISHED = 0xfffffff - 1
};
 
```
  If the rendering is finished (maximum number of samples taken), the rendering can either be reset using mi_rc_progressive_set_camera() or it can be shutdown using mi_rc_progressive_shutdown().
- Camera interaction. User-driven navigation can be implemented by calling mi_rc_progressive_set_camera() function. This call stops the current rendering, sets the camera to the specified values and restarts rendering from there. It is a full reset of the rendering and the frame number returned by mi_rc_progressive_get_update_count(), mi_rc_progressive_update_buffer() and mi_rc_progressive_update_buffer_first_subframe() are reset to 0.
```
extern "C" void mi_rc_progressive_set_camera(
    const miMatrix camera_instance,
    const miCamera*camera);
 
```
- Stop or cancel a running rendering. The rendering can be stopped or canceled at any time. Stopping mental ray kernel will complete the current frame. This can be used to allow user interaction based termination of rendering. Canceling the rendering will stop sampling immediately, not finishing the current frame, and, thus, allows to end rendering as fast as possible.
```
extern "C" void mi_rc_progressive_stop_rendering();
extern "C" void mi_rc_progressive_cancel_rendering();
```
- Shutting down the interactive mode. If the rendering has finished (e.g. of the error threshold has been fulfilled, the maximum samples were done, or the rendering was canceled), the mental ray kernel waits for a shutdown to complete. After shutting down the interactive mode, none of the API functions are allowed to be called. Interactive rendering can be restarted by calling mi_rc_run() and mi_rc_progressive_set_buffer().
```
extern "C" void mi_rc_progressive_shutdown();
```
- Tonemapping. At any time during rendering it is possible to enable or change tonemapping which can be used for applying gamma correction or adjusting the brightness on the fly. Tonemapping is implemented by registering a function which mental ray kernel will call per pixel. The tonemapping used for interactive display purpose only. The framebuffer contains the values without tonemapping. When the final image is written to disk, the desired tonemapping can be implemented by using an output shaders.
  Tonemapping functions need to map an arbitrary floating point value to the valid interval [0,1], i.e. clamping to the maximum displayable value of 1 and minimum value of 0 has to be performed. The default builtin tonemapper implements [0,1] clamping with no further pixel modifications.
  
  Although the tonemapping function can be changed at any time, the tonemap function can be called called until the current frame has been transferred to the application. Tonemapping is executed in parallel (multithreaded). The tonemapping function should be thread-safe.
```
extern "C" void mi_rc_progressive_set_tonemapper_function(
    void(*tonemapper)(float out[3], const float in[3]));
```
Image based lighting (IBL) performance has been improved. IBL compatibility with MetaSL shaders has been improved.
Finalgather combination with Irradiance particles and with Importons have been enabled. If finalgather is used in combination with Irradiance Particles, the second and higher diffuse bounces are computed with help of irradiance particles. If finalgather is used in combination with importons, importon distribution is used to drive the distribution of finalgather rays. Especially for scenes with HDR (high dynamic range) lighting, this may result in a faster convergency/better images with the same computational efforts.
DISP protocol has been extended in order to support stereoscopic images.
MetaSL 1.1 support. mental ray 3.8 will provide extended support for MetaSL shaders with the updated language specification. The stability, performance and functionality coverage are significantly improved.
MetaSL integration is subject to active development in this alpha version. In particular, the planned LLVM-based backend is not a part of this release.
Shader cache. mental ray provides a mechanism to avoid recompilation of MetaSL shaders with CIL/.Net (Windows) and C++ backend. Same mechanism applies to the C++ shaders provided as source.
If enabled (default), mental ray will create an MD5 hash code and create a .so / .dll library identified by that hash code, the platform (32 or 64 bit, Windows, Macosx or linux etc), debug or optimized, and mental ray version. This gives a unique identification of a shader and avoids possible conflicts if different version or mental ray are installed on the same system (such as 32 and 64-bit executables).

The following registries control the shading cache behavior:
- {_MI_SHADER_CACHE_ENABLE} if not set to "off" / "false", the cache is enabled.
- {_MI_SHADER_CACHE_DIR} if set, specified the directory to use for the shader cache.
- {_MI_SHADER_CACHE_SIZE} if set, specifies the size of the cache to maintain, in Megabytes.
For network rendering, mental ray distributes the source code of metasl shaders. Each host participating in the network rendering compiles shaders individually. This includes heterogeneous network support and possible combination of different backends, like CIL on windows machines and C++ on Linux machines.
mental ray 3.8 adds the ability to remove MetaSL shaders from the scene database, and to change and re-define them. This allows the workflow when the MetaSL shader source code is edited interactively. In mental ray 3.7+ it was necessary to assign a new name for such shaders.
In the current version, shader instances referring to MetaSL shader need to be deleted before MetaSL shader modification. It is planned to remove this requirement for the case when only the implementation part of the shader, but not the interface input/output parameters are changed.

Changes affecting compatibility with mental ray 3.7/3.7+.

Overview of the Information related to the changes between mental ray 3.7/3.7+ and 3.8 can be found in the upgrading38.html section of the manual.

In order to specify MetaSL backend to be used, the "{_MI_REG_METASL_BACKEND}" registry should be set to "c++","cil" or "llvm". The "{_MI_REG_METASL_CIL}" registry used in mental ray 3.7+ to enable CIL backend is ignored.
For MetaSL CIL backend, mental ray 3.8 puts most of the managed glue code into the metasl_runtime.dll assembly to be shared across all compiled shaders. This assembly file should be located in the same directory as the application and/or mental ray standalone executable. Alternatively, it should be registered in GAC ("global assembly cache") with gacutil -i metasl_runtime.dll command once at mental ray installation.
Note that default Windows security policy does not allow assemblies to be loaded from network drives. This may impose some constrains on the application / mental ray installation process.
MetaSL shaders compiled for mental ray 3.7+ cannot be used with mental ray 3.8.
On Linux and Macosx platforms, the shaders are compiled with -fvisibility=hidden flag. This is done in order to avoid symbol conflicts. Shader writers should make sure to specify the visibility attribute or use DLLEXPORT define for the exported symbols.
For versions of mental ray linked with iray, it is necessary to have the CUDA runtime library (cudart.dll/libcudart.so) to be installed on the system in order to start mental ray.

Changes in version 3.7.55.26.

Bug fixes:

Fixed a fatal if virtual or cached framebuffers were used, and the temporary directory is not writable. mental ray will print a warning and fall into in-memory framebuffers instead.
Fixed improper handling of color profiles applied to user framebuffers.
In finalgather map lookup, added a tests for shaders incorrectly flipping normal without setting miState::inv_normal properly. Fixes possible black indirect illumination rendering for some double-sided materials.

Changes in version 3.7.55.25.

Feature improvements:

To allow application-driven configuration of the external compiler tools used to compile MetaSL shaders with C++ backend, added the following registries:
- {_MI_REG_CPP_COMPILER} if set, specifies the C++ compiler
- {_MI_REG_CPP_COMPILER_FLAGS} if set, specifies the C++ compiler flags
- {_MI_REG_LINKER} if set, specifies the path to the linker
- {_MI_REG_LINKER_FLAGS} if set, specifies liker flags
- {_MI_REG_MANIFEST_TOOL} if set, specifies the manifest tool to use on Windows platforms.
mental ray standalone command line option take precedence over the registry values.

Bug fixes:

Fixed possible crash in ccmesh object tessellation.

Changes in version 3.7.55.24.

Feature improvements:

On Linux/MacOS platforms, do not spawn threads for external compiler/linker invocation if the registry {_MI_REG_DEBUG_INVOKE} is set. The change is complimentary to the environment variable added in 3.7.55.21.

Bug fixes:

For MetaSL shaders compiled with C++ backend, fixed possible crash if an array parameter in .msl file had an initializer.
For MetaSL compiler, issue a meaningful error message is a non-implemented function get_ray_data or clear_ray_data is used.

Changes in version 3.7.55.21.

Feature improvements:

Improved error reporting for mental ray standalone invoked with -mslc version for MetaSL compilation.
On Linux/MacOS platforms, do not spawn threads for external compiler/linker invocation if the environment variable MI_DEBUG_INVOKE is set. This may be useful for debugging as gdb is has issues on debugging applications using pthread fork/join.

Bug fixes:

For MetaSL CIL backend, fixed possible crash for shaders using array input parameters.

Shader changes:

For architectural material shader, fixed possible artifacts if the "thin wall" feature is been used in combination with finalgather.
For architectural material shader, fixed possible black noise artifacts in the alpha channel when trace depth is exceeded.

Changes in version 3.7.55.19.

Feature improvements:

For mental ray standalone, changed the command line option -progress_frequency <num> to accept float number for the percentage progressive message frequency. Previous versions required an integer.

Bug fixes:

For raylib integration, fixed possible crash in attaching application thread to raylib (call to mi_msg_register_thread() function) while mental ray is low on memory.
Fixed a crash if progressive rendering is used in combination with contours: disabled contour rendering. These two features are algorithmically incompatible.
For MetaSL CIL backend, fixed non-functional Light_iterator_options::set_shadows and Light_iterator_options::set_cone methods.
For MetaSL CIL backend, fixed possible invalid code generated for light samplers. Fixed possible memory corruption related to light samplers.

Shader changes:

For the architectural material round corner shader, use both (front and back) faces of the object to detect object corners instead of the face specified by global options, object or instance. For one-faces objects, some corners were not detected.
For the production mip_binaryproxy shader, fixed a crash if an invisible object (no trace/visible flags) has been created. Fixed missing database tag unpinnig which could have negative effect on mental ray database flushing when mental ray is low on memory.

Changes in version 3.7.55.16.

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 MetaSL shaders, fixed Light_iterator::set_shadows(false) option. The option is only active if the light shader is a MetaSL one.
For MetaSL CIL backend, fixed compilation of coping of Trace_options, Occlusion_options, and Irradiance_options.

Shader changes:

Fixed a crash in the architectural portal light shader if the light instance for the shader call has not been set.

Changes in version 3.7.55.14.

Feature improvements:

Support triangle vertex access for displacement shaders called on subdivision surfaces. When called from a displacement shader applied to a ccmesh subdivision surface, the mi_tri_vectors function returns the position, normal, texture, motion and derivative vertex data for an incident triangle to the current displaced vertex. For position, the last refinement level is used instead of the limit position which is overwritten during displacement mapping.

Changes in version 3.7.55.13.

Feature improvements:

For the DEBUG version of mental ray on Linux and Macosx, do not use phtread_join/fork for the external compiler and linker invocations. This allows debugging with dbgdb. gdb has a known issue (freeze) while dealing with these system function calls. Note: a disadvantage on this change is that debug mental ray will not capture the output of invoked processes.
For mental ray standalone invoked as MetaSL compiler with -mslc option, the process will return non-zero on a compilation failure.
For C++ code generated with mental ray MetaSL compiler, set the MI_RAY define. This removes the requirement to add the define on the C++ command line for shaders compiled manually.

Bug fixes:

For MetaSL compiler, fixed an issue of not passing member variables to the main() methods of shaders generated from phenomena. Also fixed possible loosing of array specifiers for such variables.
For MetaSL compiler, fixed possible crash on compiling illegal MetaSL source code, such as get_trace_depth() without parameters.
For MetaSL CIL backend, fixed possible illegal managed code due to a too low stack limit.
For MetaSL shaders, fixes possible crash on loading shaders using parameters of Shader type.

Changes in version 3.7.55.12.

Bug fixes:

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.
For MetaSL compiler, added missing support for deprecated functions like trace_immediate.
For MetaSL compiler, assure that shaders using call function will not beak C++ compilation, even though the calls will not be functional.
For MetaSL shaders, fixed texture wrong coordinate space used for texture tangents.

Changes in version 3.7.55.10.

Bug fixes:

For MetaSL shaders, fixed wrong ray origin setup used for occlusion rays.
For MetaSL XSML phenomena emulated in mental ray (precompiled MetaSL shaders mode), fixed possible wrong translation for phenomena which are not trees (example: an output of shader A is used as input for shaders B and C, and an output of shader B is used in shader C). The symptoms of the issue: API error messages on unknown shaders and broken shading due to the partial phenomena created.
For MetaSL XSML phenomena emulated in mental ray (precompiled MetaSL shaders mode), allow phenomena with color + transparency roots.

Shader changes:

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

Changes in version 3.7.55.8.

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.

Bug fixes:

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 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.
For MetaSL compiler, fixed a regression in version 3.7.55.4. For precompiled shaders, state variable modifications were ignored in some cases. In particular, bump mapping was missing in some of phenomenon.
For MetaSL compiler, fixed mental ray crash if an illegal code with input/output parameters of builtin types like Irradiance_options, which are not intended for passing between shaders. An error message is added.
For MetaSL compiler, fixed default constructor expression Type t(); and Type t = Type(); for "non-fundamental" builtin types Irradiance_options, Light_iterator_options, Sample_iterator etc.
For MetaSL compiler, fixed possible crash in the compiler for shaders with multiple s of type parameters of type Texture*.
For MetaSL CIL backend, set FLOAT/DOUBLE/HALF_MIN/MAX defines to the platform specific values.

Changes in version 3.7.55.7.

Bug fixes:

For MetaSL C++ backend, set FLOAT/DOUBLE/HALF_MIN/MAX defines to the platform specific values.
For MetaSL shaders with int2/int3/int4 and bool2/bool3/bool4 input parameters, fixed wrong initialization from .mi vector/color types.
For MetaSL compiler, fixed broken xmls phenomena connecting inputs to outputs directly.
Fixed possible MetaSL compiler crash if an (illegal) MetaSL shader in the graph misses main method.

Changes in version 3.7.55.6.

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.
For MetaSL C++ backend, fixed possible illegal code generated for the sincos function calls.

Changes in version 3.7.55.5.

Bug fixes:

For progressive rendering used with SPM-protected mental ray executables, fixed defect threshold-error-driven adaptivity.

Changes in version 3.7.55.4.

mental ray manual has been updated for this version.

Feature improvements:

The new library, intrinsics.dll / intrinsics.so, has been added. The library contains the set of 400+ MetaSL intrinsic shades. Linking this library may significantly reduce memory consumption and improve performance if MetaSL shaders are used.
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.
On Linux and Macosx, when an external compiler or linker 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. For Windows platforms, same functionality has been added earlier in version 3.7.52.14.

Bug fixes:

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.
On 32-bit MacOSX platform, for SPM-protected executable, fixed possible crash in the beginning of the second frame rendering.
Fixed echo of the scenes with IBL.
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.
For MetaSL shaders, forocclusionrays, fixed initialization of returned hit data like distance to the hit point.
For MetaSL shaders, fixed possible unknown tag fatal if an undeclared texture was used as a MetaSL shader parameter (usually passed as default value).
For MetaSL C++ backend, fixed incorrect code generated for user-defined functions having the names conflicting with standard C++ ones.
For MetaSL CIL backend, fixed incorrect code generated for the for loops containing declarations in initializers and for assignments elements of arrays of structures.
For MetaSL compiler, respect platform-specific values for the value interval defines like FLOAT_MIN/MAX, INT_MIN/MAX etc.
For MetaSL compiler, fixed possible crash at loading of phenomena containing BSDFs.

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 3.7.55.3.

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 ray-traced hair with motion blur, fixed possible rendering artifacts due to too small bounding boxes used.
For MetaSL shaders, added missing support for the light_type member of the light state.
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 C++ backend, fixed possible illegal code generated for right hand side swizzles.
For MetaSL CIL backend fixed compilation of shaders with int2, int3, bool2 and bool3 parameters.

Shader changes:

Fixed missing output passes in the "mi_car_paint_phen_x" phenomenon.

Changes in version 3.7.55.2

Bug fixes:

For hair rendering, fixed possible artifacts due to the hair radius not taken into account when bounding boxes were computed.
Fixed rendering artifacts for objects with number of motion segments higher than the global motion steps options.
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 compiler, fixed the order of handling for include files and user types declared in such files.
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.

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.

Changes in version 3.7.55.1

Feature improvements:

For the scene echo, if the value of an miBoolean is neither miTRUE nor miFALSE, echo it as an integer.

Changes in version 3.7.55.0

Feature improvements:

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:

For scenes with assemblies, fixed possible freeze and memory leak on assembly deletion at the end of the frame rendering.
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.

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 optimizes precomputation of finalgather map for animation renderings that contain a camera move. For tile rendering, the shader 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. This causes the location of finalgather points (for static object) to be consistent across the entire animation, reducing flicker. 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 erroneous 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 renaming 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_motion_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.

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.

Multi-View Images in OpenEXR files come 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 Weta Digital 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 world™, mental map™, mental earth™, mental cloud™, mental mesh®, mental™, Reality™, RealityServer®, RealityPlayer®, RealityDesigner®, MetaSL®, Meta™, Metanode®, Phenomenon™, Phenomena™, Phenomenon Creator®, Phenomenon Editor®, neuray®, iray®, DiCE™, imatter®, Shape-By-Shading®, SPM®, 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 3.9.1.37

Changes in version 3.9.1.36

Changes in version 3.9.1.35

Changes in version 3.9.1.34

Changes in version 3.9.1.33

Changes in version 3.9.1.32

Changes in version 3.9.1.31

Changes in version 3.9.1.30

Changes in version 3.9.1.29

Changes in version 3.9.1.28

Changes in version 3.9.1.27

Changes in version 3.9.1.26

Changes in version 3.9.1.25

Changes in version 3.9.1.24

Changes in version 3.9.1.23

Changes in version 3.9.1.22

Changes in version 3.9.1.19

Changes in version 3.9.1.18

Changes in version 3.9.1.17

Changes in version 3.9.1.16

Changes in version 3.9.1.15

Changes in version 3.9.1.14

Changes in version 3.9.1.13

Changes in version 3.9.1.11

Changes in version 3.9.1.10

Changes in version 3.9.1.7

Changes in version 3.9.1.5

Changes in version 3.9.1.4

Changes in version 3.9.1.3

Changes in version 3.9.1.2

Changes in version 3.9.1.1

Changes in version 3.9.1.0

Changes in version 3.9.0.14

Changes in version 3.9.0.13

Changes in version 3.9.0.12

Changes in version 3.9.0.11

Changes in version 3.9.0.10

Changes in version 3.9.0.9

Changes in version 3.9.0.8

Changes in version 3.9.0.7

Changes in version 3.9.0.6

Changes in version 3.9.0.4

Changes in version 3.9.0.3

Changes in version 3.9.0.0

Changes in version 3.8.2.0

Changes in version 3.8.1.18

Changes in version 3.8.1.17

Changes in version 3.8.1.16

Changes in version 3.8.1.15

Changes in version 3.8.1.14

Changes in version 3.8.1.13

Changes in version 3.8.1.12

Changes in version 3.8.1.11

Changes in version 3.8.1.10

Changes in version 3.8.1.9

Changes in version 3.8.1.8

Changes in version 3.8.1.7

Changes in version 3.8.1.6

Changes in version 3.8.1.4

Changes in version 3.8.1.3

Changes in version 3.8.1.2

Changes in version 3.8.1.1

Changes in version 3.8.1.0

Changes in version 3.8.0.12

Changes in version 3.8.0.11

Changes in version 3.8.0.10

Changes in version 3.8.0.9

Changes in version 3.8.0.8

Changes in version 3.8.0.7

Changes in version 3.8.0.12

Changes in version 3.8.0.11

Changes in version 3.8.0.10

Changes in version 3.8.0.9

Changes in version 3.8.0.8

Changes in version 3.8.0.7

Changes in version 3.8.0.6