mental ray Release Notes

Version 3.10.1.2

January 17, 2012

Introduction
Changes by Version
Support for OpenEXR

Introduction

These release notes list changes in mental ray since version 3.8.0.1.

Changes by Version

Note: Version 3.10 and later versions of mental ray are binary incompatible with mental ray versions 3.9.x and prior. Partial compatibility with 3.9 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 3.10 cannot be combined with version 3.9.x and earlier versions in network rendering.
iray and MetaSL plugins for mental ray 3.10 are not compatible with earlier mental ray versions.

Changes in version 3.10.1.2.

Bug fixes:

For unified sampling, fixed filtering artifacts for non-color framebuffers (such as normal and Z depth).
FIxed a crash in OEM plugin unload. The fix in version 3.10.0.18 was incomplete.

Changes in version 3.10.1.1.

Feature improvements:

For imf_copy tool, added command line option -t x [y] which allows to specify a tile size used for created tiled OpenEXR and Tiff images.
Changed the tile size for saved tiled Tiff images to 128x128.

Bug fixes:

For builtin IBL shader used with "environment lighting shadow" string option set to "solid", trace shadow against objects with shadow casting flag only.
On Linux and MacOSX platforms, fixed possible black stripes in created untiled Tiff images of small resolution (up to 320 pixels).

Shader changes:

For alembic abcimport shader, added convention to support assigning multiple materials inside of the alembic archive. If the property "material" with a string value is specified on an object, the geometry ld look up for a material with the name equal to that value (ishader would look up main scene for a material with the name equal to the property's value (in the main scene), and assign that material to the object. If the property "adskDiffuseColor" (color value) is specified, mental ray will create a lambertian BSDF with that value, put it into a new material and assign that material to the object.
For user_ibl_rect shader, light was cast to the wrong side (to the back of the light), fixed. The light was directly visible from both front and back sides which was inconsistent with the cast light, fixed.
For user_ibl_rect shader, added boolean parameter swap_uv, default value false. If set to true, the shader swaps attached texture U and V coordinates. This parameter is useful if the rectangle light has first axis along the Y direction and the second axis along the X direction, which is the case for some OEM integrations

Changes in version 3.10.1.0.

Bug fixes:

Fixed possible wrong motion vectors computation on static objects if BSP2 acceleration with enabled "trace camera motion vectors" string option was used.
Fixed a crash in multi-hosted rendering of animations with finalgatrher rebuild off option.
Fixed a memory leak for cached Tiff textures.

Shader changes:

Alembic abcimport.so and Ptex shaders had a symbol dependency causing load failure when used with integrated raylib. Fixed.
For alembic abcimport shader, fix the select parameter with value "|" to select the complete archive.

Changes in version 3.10.0.23.

Feature improvements:

Changed the tile size for saved tiled OpenEXR images to 128x128. This applies to mental ray framebuffers and to imf_copy tool. Increased tile size improves texture caching performance.

Bug fixes:

For iray rendering node, fixed compatibility with the architectural material metal.
For iray rendering mode, fixed issues in OEM tiled textures.
For iray rendering mode, fixed possible sampling artifacts in materials with multiple components (multiple layers with multiple BSDFs).
Fixed "don't echo gui" echo option (-echo omit G command line option).
For imf_disp tool, fixed possible missing animation controls when image was drag-end-dropped to the imf_disp.

Shader changes:

For shader libraries which depend on STL (stlport), i.e. usedata and useribl, remove (a large number of) STL-related symbols from the public symbol tables.
Added missing apply statements for mib_bump_map2 and contour_store_function_simple shader declarations.

Changes in version 3.10.0.22.

Bug fixes:

For iray rendering mode, fixed ground plane reflections.
For iray rendering mode, fixed possible issues in OEM noise shaders.
If unified sampling is used, framebuffers added by geometry shaders were rendered black. Fixed.

Changes in version 3.10.0.21.

Bug fixes:

Fixed possible fatal if traversal shader with string parameters was used. The issue surfaced in version 3.10.0.15.
On MacOSX platform, restored GPU acceleration of the iray rendering mode which was disabled temporarily in version 3.10.0.15.

Shader changes:

For the alembic abcimport shader, added optional string parameter "select". The value is expected to be in the form of a /-separated path into the alembic scene graph, like "/root-node/level1-nodeA". Only the selected subgraph is loaded.

Changes in version 3.10.0.20.

Bug fixes:

For iray rendering mode, fixed garbage images produced with negative camera values.
Fixed another fatal in OEM raylib integration plugin unload. This fix is a completion of the fix in version 3.10.0.18.

Changes in version 3.10.0.19.

Feature improvements:

Upgraded iray libraries from CUDA 3.2 to CUDA 4.0. On Windows platforms, iray is compiled with VisualStudio 2010, which matches mental ray build. Earlier versions of iray libraries were build with Visual Studio 2008.

Bug fixes:

For iray rendering mode, fixed OEM mirror shader support.
For iray rendering mode, fixed visibility flag support.
Fixed possible NaN values generated build builtin IBL.
Removed ifdef-out code from public headers.

Shader changes:

For user_ibl_rect light shader:
- The size of the light used in sampling was twice the expected, causing a mismatch in visible shape of the light and the lighting itself.
- When attached to rectangle light, the axes of the light were ignored and (1, 0 0) x (0, 1, 0) was assumed, which is a default in OEM integrations.
- In some cases, wrong transformation was applied to the light.
For useribl shaders (rectangle and environment), fixed possible NaN values.

Changes in version 3.10.0.18.

Bug fixes:

Fixed possible fatal in OEM raylib integration plugin unload (incomplete, completed in version 3.10.1.2). This is an equivalent of the fix in version 3.10.0.16.
For imf_disp tool, animation control tools were sometimes not displayed. Fixed.

Shader changes:

Fixed compiler warnings in base shaders with public source code.

Changes in version 3.10.0.17.

Bug fixes:

Fixed a moderate per frame memory leak. A significant amount was accumulated in animations and swatches rendering.

Changes in version 3.10.0.16.

Feature improvements:

For scenes with displacement, significantly reduced negative impact of unnecessary large max displacement values set multiple times higher than the actual maximal displacement. For scenes with extreme discrepancy, a total rendering speedup of more than factor two has been observed.
For iray rendering mode, added support for OEM color-by-object shader.
Added builtin_bsdf.mi file to the releases. The file is not needed by mental ray, but may be used to lookup shader declarations manually or by tools integrating mental ray library.
For imf_copy tool, added support for -r command line option (tiled/non-tiled output) for Tiff output files.

Bug fixes:

Fixed possible freeze or unknown tag fatal at raylib exit. Regression in version 3.10.0.15.
Changed mi_api_registry_eval() and mi_api_registry_lookup() functions to return char * pointer instead of const char * pointer. The caller of these functions is responsible for releasing (with mi_mem_release()) of the returned string. Changed constness makes this more intuitive.
Fixed several minor memory leaks in registry lookups.
Removed unused automatic creation of GUI attributes for phenomena declarations. The attributes were not removed before raylib exit, which could lead to unnecessary memory consumption raising on interactive work.
Fixed parsing of default values fordata declarations. On user data, the default value was applied to the wrong parameter (one before the current).
Fixed possible crash when loading existing finalgather map file with finalgather rebuild off option.

Changes in version 3.10.0.15.

Known limitations in this version:

On MacOSX, iray rendering runs in CPU mode only.

Feature improvements:

For iray rendering mode, added support for matte objects rendering.

Bug fixes:

For hair ray tracing, fixed possible crash during hair acceleration construction. Precision issue.
When writing out multiple images to the same OpenEXR file, fixed possible wrong color profile applied. The color profile specified for the first image was applied to all images saved to the same file.
Fixed database and memory leaks for shaders and phenomena using string and/or unnamed texture parameters. The latter in particular fixes a major memory leak in subsurface scattering phenomena which do not specify lightmap and depthmap texture parameters.
Fixed a database tag and minor memory leak when a local or writable texture was used.
For iray rendering mode, fixed possible artifacts if two neighbor volumes have identical IOR.
For iray rendering mode used interactively, fixed possible artifacts in texture re-baking.
For imf_disp tool on MacOSX, fixed drag-and-drop functionality.
If output image format type is not specified, write out files with .sxr extensions as OpenEXR.
When writing out multiple images to the same OpenEXR file, added missing info message on writing the second and subsequent framebuffers.
Fixed formatting of messages for debug "db dump" command in .mi files.
For memory statistics, report iray memory usage under "IRAY" instead of "RC".

Shader changes:

Fixed possible artifacts in environment sampling for user_ibl_env shader.
For userdata shaders, avoid possible fatal for misconfigured scenes.

Changes in version 3.10.0.14.

Feature improvements:

For iray rendering mode, fixed possible shading artifacts for scenes rendered with a lot of computational efforts.

Bug fixes:

Fixed echo of shader instances with struct parameters, where values of each of members of the struct is equal to the default value.
Fixed possible failure to update textures between frames. If several API texture names refer to the same file on the disk, and the textures are opened with the same parameters (like filter), mental ray shares the content of the textures in memory. If such shared textures were modified on the disk between frames, mental ray did not reload them even if incremental API updates were done. With this fix, mental ray takes into account the timestamp of the file into when a decision to share textures is done.
For iray rendering mode, added support for mib_color_mix used as a material shader.
For iray rendering mode, fixed possible concentric artifacts in wave maps.
For imf_disp tool, fixed a crash when exporting tiled OpenEXR file (-p or -r command line option, for OpenEXR -p (mipmap pyramids) implies -r (tiles)).

Changes in version 3.10.0.13.

Bug fixes:

When a registry lookup with "include" attributes triggered parsing of .mi files with (nested) include statements, the file order was broken and some of includes were possibly not processed. Fixed.
Nested include files had a depth limit of 9. The limit is raised to 4096 (and kept to just break possible self-include infinite recursion).
For OpenEXR images, B44 compression was used instead of PXR24. Fixed.
Fixed possible rare multi-threading scalability issue: for simple scenes, some threads could stop participating render in rendering.
For imf_disp tool, improved halftone image mask.

Changes in version 3.10.0.11.

Bug fixes:

Fixed a shadow tolerance issue for directional lights if BSP2 acceleration is used.
Switched the default format for TIFF and EXR output images back to scanline, the tiled format can still be specified with an explicit setting in the framebuffers options.
For iray rendering mode, fixed some numerical imprecision that happened in rare corner cases.
For iray rendering mode, fixed an issue with motion blur that caused light intensities to flicker between frames.

Changes in version 3.10.0.10.

Bug fixes:

Fixed shadow artifacts for one-sided surfaces if BSP2 acceleration is used.
For iray rendering mode, fixed possible metal material visual differences vs. iray 1.2 version used with mental ray 3.9.1.

Changes in version 3.10.0.9.

Bug fixes:

For mi_tri_vectors() function, fixed lookup for texture derivatives (bumps) and for derivs. In particular, this fixes anisotropic shading artifacts.

Changes in version 3.10.0.8.

New features:

Added support for import of Alembic files. A new shader library (abcimport.dll/.mi) containing a single geometry shader
```
    declare shader geometry "abcimport" (
        string  "filename",
        scalar  "time"      default -1.0,
        boolean "assembly"  default off 
        )
        version 1
        apply   geometry
    end declare
```
is added. The shader searches for the Alembic file specified by the "filename" parameter using the assemblies search path set (-assembly_path command line option).

If "time" parameter is non-negative, the geometry for that time is selected. If it is negative, mental ray's miCamera::frame_time is used (default).

If "assembly" parameter is set to on, the geometry shader itself locates .abc file, computes bounding boxes and creates a procedural assembly. The actual geometry (including possible hierarchy) is generated on demand when the bounding box is hit. If "assembly" parameter is set to off, geometry is loaded immediately during scene preprocessing. This is the default mode.

Materials, approximations and flags are inherited from the instance the geometry shader is attached to.

Alembic-related libraries are linked statically into the shader DLL's, the shader does not have dependencies on other DLLs.

Bug fixes:

For iray rendering mode, fixed artifacts if IOR of 1.0 was used.
For mental ray legacy framebuffer API, such as mi_output_image_open() function, fixed the selection of the color framebuffer assuring this is the primary one.
In scene preprocessing, fixed inheritance of override approximations into assemblies.
Avoid a crash if a discontinued builtin shader (such as builtin_constant_color) is used in the scene.
For imf_disp tool, fixed possible crash on Linux platforms, in particular if remote display is used. Fixed GUI controls and frame number display in the image pipe mode.

Shader changes:

Removed obsolete platform defines for shaders provided with source code.

Changes in version 3.10.0.7.

Feature improvements:

For iray rendering mode, added support for round corners feature.
For imf_disp tool, restored image pipe functionality known from imf_disp shipped with mental ray 3.7 and prior (command line ray .... -imgpipe | imf_disp - ).
For imf_disp tool used to display animation rendering, avoid jumping back to display the first frame rendering (or finalgather preview of the first frame rendering) after the animation ends.
Added support for camera near/far clipping planes in combination with scanline off option. If the string option� "trace camera clip" is set to true, the near and far clipping planes are respected for primary rays shot from the camera. This string option should be used with care in combination with lens shaders benting eye ray directions (for example, fisheye-style shaders) as it may lead to unexpected results: clipping is done based on the camera only. The default value for� "trace camera clip" string option is false (compatible with mental ray 3.9 and prior).
Geometry shader api is extended with function char * mi_api_assembly_path_search(char *filename). A geometry shader may use this function to determine the absolute file name according to the set of rules which mental ray kernel uses to search for assembly .mi files. The argument filename should be mi_mem_strdup()-ed, and the returned result should be mi_mem_release()-ed.

Bug fixes:

For BSP2 acceleration, fixed possible NaN artifacts for scenes containing multiply-instanced objects with some moving and some static instances.
For BSP2 acceleration, improved precision of shadow rays. Internally, the direction of shadow rays traced for directional lights is inverses from light-to-object (old) to object-to-light (new). The shader API is unchanged.

Fixed possible crash if legacy finalgather maps were used.
For imf_disp tools used to display sequences of frames, changed File->Save As menu to save the frame currently displayed instead of the first frame of the sequence.
In rasterizer, miState::importance was not properly re-initialized between samples, fixed.
For iray rendering mode, fixed possible crashes for scenes with both portal lights and normal lights.
For iray rendering mode, fixed possible gamma correction artifacts.
For iray rendering mode, fixed possible crash on 32-bit Windows platform.

Shader changes:

The ptex.dll shader library (mib_ptex_lookup shader) is changed to link the standard Ptex library statically. This is done in order to avoid possible conflicts with other version of libPtex.dll possibly installed on the system or used by an application integrating raylib.
For architectural portal light shader, downgraded error message to warning.

Changes in version 3.10.0.6.

Feature improvements:

For iray rendering mode, added support for camera window.
For iray rendering mode, added more and improved support for OEM texture shaders.
Improved memory usage reporting messages.

Bug fixes:

For SPM-protected version of mental ray on Windows machines, the freeze observed in version 3.10.0.5 was resolved.
On Linux and MacOSX platforms, fixed possible broken phen/shading tree evaluation (gcc compiler bug). In particular, this fixes possible displacement artifacts.
On multi-CPU machines, improved performance of the scene rendering startup.
Fixed possible rare crash usually observed at the beginning of rendering stages (photon emission, finalgather precomputation or tile rendering).
For multi-hosted rendering, in some cases slave/satellite machines did not receive rendering tasks, which lead to their low participation. Fixed.
Geometry secondary root shaders of material phenomena were not executed, fixed.
Fixed possible hair shadow artifacts.
Added missing gradients (or, equivalently: spherical harmonics) for finalgather points computed with importance weights (see "finalgather importance" string options feature of mental ray 3.10)).
For finalgather, fixed possible lookup slowdown due to improperly build tail of the paged map. Regression in 3.10.0.5.
For imf_disp tool, fixed scrambled channels for OpenEXR image layers created exported by mental ray from normal and motion framebuffers.
For MetaSL shaders, fixed possible missing texture default initialization.
For MetaSL compiler, fixed compilation of the compound shader example from MetaSL documentation.
For MetaSL compiler C++ backend, fixed a regression in compilation of BSDF nodes.
For iray rendering mode, fixed support for "reverse" parameter of the mia_skyportal shader.
For iray rendering mode, fixed a regression in mia_photometric_light light distribution.
For iray rendering mode, fixed possible artifacts due to IOR computation on transparent materials.
For iray rendering mode, fixed possible artifacts after updating scenes with emissive materials.
For iray rendering mode, fixed possible artifacts for cylinder lights.
For iray rendering, fixed possible artifacts in the initial subframes due to a failure to re-render fragment after a driver timeout was detected.
For iray rendering mode used interactively, fixed possible failure to update objects created by geometry shaders.
For progressive rendering and iray rendering mode used interactively, added a synchronization in order to avoid possible race condition issues on scene modification. In order to modify scene, applications editing scene need to pause rendering. The changes are picked up by renderer after the rendering is resumed.
For progressive rendering and iray rendering mode, fixed possible crash on interactive camera updates (regression).
For mental ray standalone running on MacOSX version 10.6, fixed a crash if iray rendering mode was used. CUDA runtime library used by iray depends on CoreFoundation. Loading CoreFoundation library from a thread which is not the main thread of application leads to a crash on MacOSX 10.6. In order to resolve the issue, mental ray standalone on MacOSX is now linked against CoreFoundation library. For integrators: applications using mental ray library using iray rendering mode should assure that CoreFoundation is linked.
Fixed several bugs in the connectivity to the cloud.

Shader changes:

On Windows platforms, fixed possible broken progress/info/error/debug messages generated by shaders if the messages contained percentage (%) sign. shader.lib call forwarding did not encapsulate strings properly. In particular, the fix applies to the subsurface scattering shader.
For the production mip_motion_vector shader, added support for the string option "trace camera motion vectors" introduced in version 3.10.0.5. The camera motion is no longer added in the shader for the scanline off + BSP2 acceleration case.
For fast subsurface scattering shaders, added support for camera offset.
For mia_photometric_light shader, fixed possible crash if the shader was used on a light instance with non-orthonormal transformation.

Changes in version 3.10.0.5.

Known limitation in this version:

SPM-protected mental ray standalone and library may freeze on some Windows machines.

Features affecting compatibility:

The scanline on option is deprecated. The default value is off. With scanline off option, was camera motion is not taking into account when computing miState::motion and motion framebuffer. Now, if BSP2 acceleration is enabled in combination with scanline off option, the string option "trace camera motion vectors" (default: off) can be used to enforce adding camera motion effect to the computed motion vectors.

Feature improvements:

Allow to specify min / max samples per instance with the following syntax:
```
     instance ...
        ...
        samples <float min> <float max>
        ...
     end instance
```
Added min/max samples inheritance the scene graph.
Added bicubic texture filtering lookup. An overloaded lookup_filter_color_texture() method is added to mi::shader::Interface, which takes filter type (bilinear/bicubic) as a last argument. Example of the usage can be found in the mib_texture_filter_lookup_2 shader source code.
Improved performance of finalgather lookup for scenes with glossy reflections.
Improved hair rendering performance.
Added support for stereoscopic OpenEXR ("sxr") files. Rendered images are saved in "sxr" , if framebuffer's type is "sxr", or if the registry's "{_MI_REG_STEREO_EXR_MULTIVIEW}" value is set to "on".
Known limitation: imf_disp is able to display only mental ray-generated SXR files. For arbitrary ones, camera names may be interpreted as conventional layer names.
For integrated MetaSL compiler, improve performance of the generated code by means of static code analysis. This may significantly slowdown the compilation phase.
For scenes with many database elements running low on memory, improved scalability by decreasing the time interval of the database being locked.
Improved comment printed out in the beginning of echoed .mi files.
For iray rendering mode, added progress reporting for texture baking.

Bug fixes:

Particle map objects are not supported for bsp acceleration. If such an object is detected, an info message is printed and bsp2 acceleration is used instead.
Fixed possible memory corruption for some shaders with small size outputs (less than 16 bytes) connected as inputs to other shaders. The corruption occurred if such shader assumed that it is safe to fill the result with a color.
For MetaSL shaders, fixed possible update of the shader bytecode if a MetaSL file with the code definition was reloaded.
For MetaSL shaders using direct_light()/indirect_light() functions, fixed artifacts due to a wrong argument order interpretation.
For iray rendering mode, fixed a crash if irradiance particles option was enabled (irradiance particles are not used in iray rendering mode).
For iray rendering mode used interactively, fixed out-of-memory crash if a light shader is updated incrementally.
For iray rendering mode used interactively, fixed a crash if object material's was changed incrementally to a null tag.
For iray rendering mode used interactively, fixed possible artifacts in rendering scenes after updates.
For iray rendering mode used interactively, fixed possible crash on incremental shader/material change which disables a texture.
For mental ray progressive rendering and iray rendering mode used interactively, fixed possible deadlock freeze. The deadlock was in particular possible if rendering was cancelled during the scene rendering initialization.
For raylib integrations using progressive rendering or iray rendering mode, fixed the return value of mi_rc_progressive_get_update_count().
Fixed several issues in the cloud rendering connectivity.

Shader changes:

The mib_texture_filter_lookup_2 base shader get new parameter allowing to specify bicubic filtering.
Fixed possible artifacts for mia_photometric_light shader on visible lights.

Changes in version 3.10.0.4

Feature improvements:

iray v2 plugin interface has been changed to enable both-way compatibility in the future. Compatibility with older revisions of iray v2 and iray v1.2 is not maintained.
Adjusted message reporting: if the "memory" verbosity bit is set, each printed line contains the approximate amount of physical memory used by mental ray on the moment the message is printed. The amount is only an approximation due to the asynchronous message printing and asynchronous memory statistics collection.
For ccmesh objects, improved texture seam computation for multiple texture spaces. Up to four texture spaces are supported. Earlier version took up to two spaces into account.
For ptex texture support, added a registry key {_MI_REG_CCMESH_PTEX}. If its value is set to "on", all ccmesh objects are tessellated with ptex support capability, even if the bary flag is not specified on such objects.
Improved performance for ray-traced shadows on hairs (10-30%).
For ray-traced hairs, improved image quality by increasing precision of intersection calculation.
For ray-traced hair rendering, added a transparency depth limit of 250, similar to the maximum depth limit used for hair rendering in rasterizer.
Improved infrastructure for cloud rendering support. Due to changes interfaces, the functionality to read images from URLs (instead of files) is no longer needed and is removed.
Improved texture filtering for MetaSL shaders with texture remapping.

Bug fixes:

For map containers (including finalgather map files), a temporary file was created in the current directory, causing an error if the directory was read-only. This could in particular be the case if the application or mental ray standalone was started with a script changing local directory, or with a Windows shortcut specifying "Start In" directory.
Allow finalgather legacy mode ("finalgather legacy" on string option) to be used in combination with the new finalgather map format introduced in mental ray 3.9. In legacy mode, mental ray 3.8 containers are used for finalgather maps, which are known to deliver better performance for some scenes with a combination of finalgathering, glossy reflections, and moderate finalgather map size.
For finalgather maps appended to the file for animation rendering, some of the points appended were not taken into account during rendering. Fixed.
Fixed possible fatal error if additional framebuffers were added by a geometry shader.
For texture lookups based on differentials (like mib_texture_filter_lookup_2 shader), the lookup results were too blurry. Fixed.
Improved handling of ccmesh objects which are non-2-manifolds. Added a control to avoid possible infinite loops. In addition, added a registry {_MI_REG_CCMESH_CHECK_TOPOLOGY}. If set to value "on", an expensive topology tests is done.
For ccmesh objects, fixed possible crash for a corner located on a boundary.
For iray rendering mode, fixed artifacts if double sided materials were used (regression).
For debug executables, fixed an assert if scanline on rendering was used in combination with unified sampling.

Shader changes:

For ptex shader library (mib_ptex_lookup shader), search for ptex texture files using the same texture path search mechanism as used by the kernel for traditional textures.
For ptex shader library (mib_ptex_lookup shader), fixed handling of 4-channel textures.
For the subsurface scattering shader, added two skin phenomena misss_phen_skin2_mia_phen(_d) based on the misss_fast_shader2(_x) added in version 3.10.0.3. The shader reference documentation is extended with these new shaders.

Changes in version 3.10.0.3 (incomplete list).

Known limitation in this release:

On windows platforms, iray runs in CPU mode only. GPU support is pending on Visual Studio 2010 by CUDA compiler. GPU support is restored in version 3.10.0.5.

New features:

Ptex texture support for ccmesh objects added.

Lookup to ptex textures is implemented in the shader mib_ptex_lookup:

    declare shader color "mib_ptex_lookup" (
        string  "filename",
        scalar  "width"             default 1.0,
        scalar  "blur"              default 0.0,
        integer "filetype"          default 4,
        boolean "interpolate_level" default off,
        scalar  "filter_sharpness"  default 1.0)
        version 1
    end declare

In the current version, the shader is compiled into the dynamic library called ptex.dll/so. This library depends on the libPtex.so/dll, which contains the Ptex library itself.
.mi syntax has been extended with the "bary" keyword which can be added to a ccmesh object (before the data block). If present, ccmesh tessellation generates barycentric and face index information which can be retrieved with mi_query function.

The new mi_query mode miQ_PRI_SDS_UV can be used in order to obtain ccmesh face UV coordinates. The value stored into a parameter is a float pointer to an array of 7 values. The values 0+1, 2+3, 4+5 contain barycentric coordinates of the vertices of the triangle hit. The value #6 contains the index of the face (in unsubdivided representation), as a float. The storage used to keep the seven float values is in the internal cache, so a copy of the value should be kept if used after a subsequent mi_query call, tracing rays or calling other shaders.

Added new user light shaders for efficient importance-driven light sampling:

    declare shader
    color "user_ibl_env" (
        color texture "texture",               # environment lookup shader or texture
        integer       "samples" default 0,     # limit number of arealight samples
        color         "color" default 1 1 1,   # light color scaling
        scalar        "intensity" default 1,   # scalar multiplier
        integer       "shadow_mode" default 2, # 0=no shadow, 1=opaque, 2=transp
        scalar        "rotate",                # rotation
        boolean       "as_reflection" default false, # treat the texture as a reflection
        boolean       "primary" default true
    )
    apply light, environment
    version 3
    end declare

This shader is similar to the builtin area light, but provides extended parameter set.

    declare shader
    color "user_ibl_rect" (
        color texture "texture",               # environment lookup shader or texture
        integer       "samples" default 0,     # limit number of arealight samples
        color         "color" default 1 1 1,   # light color scaling
        scalar        "intensity" default 1,   # scalar multiplier
        integer       "shadow_mode" default 2, # 0=no shadow, 1=opaque, 2=transp
        scalar        "cone" default 90.,
        scalar        "falloff" default 2.0,
        scalar        "angular_falloff" default 1
    )
    apply light
    version 3
    end declare

This shader implements texture-driven importance light sampling on a rectangle light.

The shaders are a part of the new useribl.so / useribl.mi library.

Feature improvements:

Added support for names/types for texture and user spaces on vertices.
- Extended .mi syntax in order to support names/types for texture and user spaces on vertices.
  The vertex info follows the syntax:
  
  vertex
  
  p | n | d | d2 | m <opt_size> | b <opt_size> | t <opt_size> <type> <opt_name> ... <opt_size> <type><opt_name> | t <opt_size> <type> <opt_name> ... <opt_size><type> <opt_name>
  
  where type is scalar, vector, color or integer. An example of a trilist object syntax would be:
  
  trilist
  vertex 4 p n t 3 scalar "mytexture" u 1 vector "diffuse"
  triangle 2 [...][...]
  end trilist
- Added the following mi_query() modes in order to support named per-vertex data:
  - miQ_PRI_VERTEX_INFO: returns the information of a named vertex data element. The parameter to return is caller-owned miApi_vertexinfo structure. The specific selector used in the const char * name string. Note that the name ins the miApi_vertexinfo is valid until the next mi_query() call only. Currently, the query mode is supported for trilists, polygon and ccmesh objects only.
  - miQ_PRI_VERTEX_INFO_INDEX: returns the information of a vertex data element specified by index. Same as miQ_PRI_VERTEX_INFO, except that the selector parameter is an unsigned integer instead of the string.
  - miQ_PRI_VERTEX_INFOS: returns the number of named vertex elements, to be used as an upper limit to the selector of the miQ_PRI_VERTEX_INFO_INDEX query mode.
- For integrations, added function mi_api_vertex_info_add() which specified a name/type of a single vertex element.

Added support for per-face user data for ccmesh object.
- The syntax for ccmesh objects has been extended, like in the example below:
  ccmesh "s" polygon 1 vertex 6 data 1 3 p 0 1 2 3 4 5 6 data [1234][0 1 2] end ccmesh
  
  The first data statements specifies the number of arbitrary ints and the number of ints per face. The second provides the arbitrary data followed by the per polygon data.
- Added the following mi_query() modes in order to improve support for per-face data:
  - miQ_PRI_NUM_DATA: returns per-face data size and layout information. Caller-owned pointer to the miPri_data structures should be passed as the mi_query parameter. The parameter is filled with the data. Sizes are counted in 32-bit integers/floats. The query call is supported for rendering stage shaders and for displacement shaders.
  - miQ_NUM_USERS: returns the number of user user vectors per vertex.
  - miQ_USER_DIM: return the dimension (number of scalars) of user vector. The user vector is selected based on the additional selector parameter passed to mi_query(), with the valid range from zero to the return of miQ_NUM_USERS query mode.

Added the mi_query() mode miQ_PRI_UV_DIFFERENTIALS. In order to use this mode, computations of ray differentials should be enabled ("ray differentials" true string option). The parameter passed should be a caller-owned array of 4 floats. On success, the 4 floats are filled with dx, duv/dx, dy, duv/dy geometric differentials.
Refurbished finalgather + irradiance particles combination in order to allow irradiance particles to drive finalgather importance sampling.
Added string option "finalgather importance", with the following possible values:
- "off": If irradiance particles are on, their contribution is used for the computation of the secondary diffuse component. Irradiance particles are not applied to drive importance sampling for finalgather rays
- "partial": the importance-based mechanism of irradiance particles is used to drive finalgather ray distribution, and irradiance particles are used for secondary diffuse bounces.
- "full" or "on" : like "partial", and in addition for shading of finalgather rays, light sampling loops would return black and the direct diffuse illumination is obtained from the irradiance point maps. Note that in this mode, the glossy component of finalgather ray hits is suppressed, which may read to slightly darker images.
In addition, finalgather can be used with importons without irradiance particles. In that case, importon map drives finalgather ray distribution.
Experimental: for finalgathering, added an experimental finalgather lookup mode which tests the visibility from the lookup point to the FG interpolation point candidates. The mode can be enabled with the "finalgather points test" string option, or with the new miIrrad_options::finalgather_points_test value set to 1.
For builtin ambient occlusion, test the hits against shadow cast object instead of "any flag set". For shader calls, miAmboccl_options::flag, which allows to specify a bitmask of multiple flags (shadow/reflection/refraction/...) to be used in the ambient occlusion tests. The default one is shadow.

Bug fixes:

For ccmesh objects, fixed possible texture seam artifacts.
Fixed possible very rare crash due to NaN direction generated for diffuse or glossy reflections.
Fixed possible tile border artifacts if coverage framebuffer is used in combination with motion blur.
Fixed possible unknown tag fatals at the end of rendering. Regression in version 3.9.4.5.

Shader changes:

Added new subsurface scattering shader misss_fast_shader2(_x) with features extended for production purposes.

Changes in version 3.9.4.7

Bug fixes:

Fixed unknown tag fatals and crashes in iray rendering mode introduced with texture filtering cache. Regression in version 3.9.4.5.

Changes in version 3.9.4.6

Feature improvements:

For iray rendering mode, improved performance of baked textures caching by using finer caching selection.
For cloud rendering, implemented efficient mechanism to echo changed tags only.

Bug fixes:

With finalgather rebuild off option, the finalgather points from the existing finalgather map file(s) or previous frame(s) were not used when detecting if new finalgather points need to refine the map. This lead to unnecessary computations of new points and exposing of the finalgather map file size for long animations. Regression in mental ray 3.9. Fixed.
For iray texture baking introduced in version 3.9.4.5, fixed possible crash if scene is changed during baking process.
Allow phenomena root declarations to be of any type. This is needed to support MetaSL phenomena in the "emulation" mode ({_MI_REG_METASL_EMULATE_MI_PHEN} registry).
For MetaSL compiler, fixed a crash in too long swizzles (like a.xxxxx). Issue an error instead.

Changes in version 3.9.4.5

Feature improvements:

For iray rendering mode, added support for caching baked textures across multiple frames with incremental scene changes.

Bug fixes:

In progressive rendering, fixed a crash when integrating application decides to reinherit lights between frames.
For displacement, reduced the number of warnings on improperly set max displacement to one per object.
Assure that shader libraries specified on the mental ray standalone command line are loaded in the order of specification. For network rendering, also assure the loading order on slaves/satellites receiving the list of shader libraries from the master host.
For MetaSL compiler, fixed possible "name already used" error if multiple shader use external names.
For MetaSL C++ backend, fixed incorrect C++ code generated for global struct members.

Changes in version 3.9.4.4

This version cannot be combined with mental ray 3.9.4.3 and prior in network rendering.

Feature improvements:

For ccmesh objects, added support for the approximation sharp parameters (smooth vs. facetted shading normals).

Bug fixes:

For BSP2 acceleration when used for motion blur scenes: fixed possible shading artifacts for scenes which contain multiply-instanced objects and some static objects.
For ccmesh objects with unsupported approximation types (warning situation), avoid a fallback into using too fine tessellation. The coarse one based on min subdiv level is used instead.
For unified sampling, the per-object min/max sample fix in version 3.9.4.0 was incomplete. Fixed the support for per object max samples limit.

Changes in version 3.9.4.3

Changes affecting compatibility:

Iray integration interface version has changed to 8. mental ray 3.9.4.3 will not work with iray plugin from earlier versions.

Bug fixes:

Fixed a raw (single pixel) texture lookup if the texture filter value is set to -1.
Added missing sample diagnostic mode for progressive rendering. Only number of samples is supported.
For progressive rendering, do not stop rendering tiles with no object hits due to possible non-geometric effects.
For MetaSL compiler LLVM backend, fixed black images rendered with BSDF shaders. Regression in 3.9.4.2.

Changes in version 3.9.4.2

Bug fixes:

Fixed possible stack overflow crash for scenes with large number of hull objects (object with materials with no surface shader).
On 64-bit platforms, non-SSE version of texture filtering and photon map lookup was executed. Switched code to the SSE path, which leads to a minor performance improvement.
For MetaSL compiler, fixed possible compiler crash in handling Trace_options.

Changes in version 3.9.4.1

Bug fixes:

If mental ray standalone was used with pipe input (like cat file > ray ....), the parser buffer could get corrupted after parsing the first render command, i.e. only the rendering of one frame was possible. Fixed. Note: for a continuous pipe stream it is still necessary to push some data / new lines in order to fill input buffers and trigger the parser to process the input. Appending 8K of new lines is recommended.

Shader changes:

For architectural material library, added a new shader mia_exposure_photographic_rev, which approximately inverses the effect of mia_exposure_photographic shader. The parameter sets of those two shaders are identical.

Changes in version 3.9.4.0

Feature improvements:

For raylib integration, provided initial implementation of the Cloud connectivity support.
For mental ray standalone, added new command line option:
- -unified_sampling <on/off> enables/disables the unified sampling.
- -unified sampling <min> <max> [<quality> [<error_cutoff>]] enables unified sampling and sets the min/max samples parameters, as well as optional quality and error cutoff.

Significantly improved performance of displacement presampling.
Improve imf_disp's performance for large resolution images displayed during rendering.

Bug fixes:

With "contrast all buffers" string option set, contrast tests were not applied in several first user color framebuffers. Regression in mental ray 3.9, fixed. This includes the wrong offset fix from version 3.9.1.39 on the mental ray 3.9.1 branch.
For BSP2 acceleration used on scenes with large bounding box or distant lights, fixed possible ray-traced shadow precision artifacts.
On Linux and MacOSX platforms, fixed broken decompression of OpenEXR images using PIZ compression. In most of cases, the alpha channel was wrong.
For Tiff images with tiled mipmap pyramids, fixed possible artifacts due to improper loading (texture caching only) of upper levels with tile resolution smaller than the level resolution.
Fixed distribution used in the mi_reflection_dir_anisglossy_x() function (anisotropic glossy reflection). Regression in mental ray 3.9.
Fixed error messages if a primary framebuffer is not of a color or alpha/intensity type. Such framebuffer is now downgraded to non-primary one like in mental ray 3.8 and prior.
For progressive rendering, use adaptivity sampling criteria which leads to a better match of images to those rendered with unified sampling in non-progressive mode.
For unified sampling / progressive rendering, clip per-object min/max samples to the maximal value of 127. The internal storage is limited to one byte.
For raylib integrations using progressive rendering or iray rendering mode in combination with application framebuffer, fixed possible crash on the camera resolution change.
Fixed possible virtual memory leak if a scene with multiple memory-mapped framebuffers was rendered with progressive rendering.
Slightly improved scalability on multithreaded machines by reducing locking in the scheduler.
For multihosted rendering, avoid redundant communication on scene changes, in particular when animations are rendered.
For MetaSL compiler, fixed possible illumination artifacts. Expressions like a/(b*c) were interpreted as a/b*c on the compiler code path used by mental ray.
For MetaSL LLVM backend, fixed possible runtime crash if BSDFs are used.
For MetaSL LLVM backend, avoid installing signal handlers in LLVM. This could have negative affect on applications integrating raylib and relying on their own signal handling mechanism.
For MetaSL compiler, fixed several issues in SSA analysis.
For MetaSL compiler, fixed several memory leaks.
For iray rendering mode, converted several info messages to progress one for a better verbosity match.

Shader changes:

For mia_physicalsky/mia_physicalsun and mip_rayswitch_environment shaders, fixed background visible through objects with hull materials. Hull materials are frequently used to enable volumetric effects.

Changes in version 3.9.2.13

Bug fixes:

Tiled TIFF output image files were broken, regression in version 3.9.2.11. Fixed.
Fixed some memory leaks for MetaSL compiler.
For MetaSL compiler, fixed possible crash when coping nested phenomena or when local variables were used inside of if/else statement.

Changes in version 3.9.2.12

Bug fixes:

For ccmesh objects, fixed a crash in reallocation with object containing creases.
For MetaSL LLVM backend, fixed a crash if a shader name ends with "_bsdf".

Changes in version 3.9.2.11

Feature improvements:

Improved quality of high-depth detail shadow maps. A new string option: "detail shadowmap contrast", default value 0 0 0, is added. If set, its value is used to combine multiple steps in the visibility drops as long as the difference in the visibility does not exceed the option value. With moderate positive values (for example, 0.15 0.15 0.15), a much depth-deeper falloff of the visibility would be typically stored in a more compact form than the one used in earlier versions. The accuracy parameter for detail shadow maps is deprecated and is no longer used. The shadowmap bias parameter is respected. If not set, the value computed based on the scene box is used.
The "tiled" attribute syntax for framebuffers (added in version 3.9.2.8) is supported for Tiff image files. Like for OpenEXR files, tiled images are written out by default.
Avoid printing an API error message if an instance group is used as a shader/user data light parameter.

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.
For the .mi parser, the namespace was not restored on an error of a shader re-declaration. The rest of the scene was parsed into the namespace of the shader. Regression in version 3.9.1.24. Fixed.
For progressive rendering, fixed possible tile artifacts, in particular for low sampling rates and sharp edges. Tiles with low percentage of pixels which need oversampling were skipped completely, which caused a mismatch in sampling rates across the tiles.
For some multicore machines, the scalability of finalgather precomputation stage in the "finalgather precomp legacy" mode (mental ray 3.7 and prior compatibility) was poor. Improved.
The mi_db_deferredflush() function for propagating of multiple database tag changes to other hosts in multihosted rendering was not thread-safe. Fixed.
Fixed a minor memory leak when PNG image file was created.
For iray rendering mode, fixed possible precision artifacts when rendering background plane.
Fixed a memory leak (unused texture spaces data) in iray rendering mode.
Fixed several memory leaks in MetaSL compiler.
For MetaSL compiler, fixed compilation if shaders returning unsupported array and string types.
For MetaSL LLVM backend, fixed possible crash when implicit type casts (like float3 x = 0;) were used.

Changes in version 3.9.2.10

Note: for network rendering, this version of mental ray cannot be combined with mental ray version 3.9.2.9 and prior.

Feature improvements:

Increased the number of database tags which can be held simultaneously to approximately 1 billion (10^9). Earlier mental ray version allowed approximately 16 Mio of tags. This change introduces a network rendering incompatibility with earlier versions. Significantly improved performance for creating of large amount of tags.

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 3.9.2.8.
Fixed wrong ray type visibility flags used for mi_trace_probe_x() function in combination with BSP2 acceleration.
Fixed wrong gamma correction offset if sRGB (mental ray name: "micsSRGBg") color space was used.
For subdivision surface objects with displacement (legacy implementation, HSDS), fixed wrong warning on the amount of displacement.
For raylib integrations using progressive rendering and iray rendering mode, fixed possible freeze of a fast sequence of stop/resume rendering events.
For MetaSL compiler, fixed possible crash.
For MetaSL compiler, fixed several memory leaks.

Changes in version 3.9.2.8

New features:

Added two new standalone tools to deal with the map containers:
- The map_info utility prints out the information about main container.
  Syntax:
  
  map_info <map_file>
- The map_copy utility allows to convert maps between ascii (follows .mi syntax), open binary (simplified, no lookup acceleration structures) and close binary (containing proprietary lookup acceleration structure formats).
  Syntax:
  
  map_copy [opts] map1 ... mapN out_map
  
  Options:
  
  -h: print help information
  
  -o cb|ob|oa: output format to use, closed binary (default), open binary or open ASCII
  
  -v: verbose output

Feature improvements:

Framebuffer syntax has been extended in order to allow custom attributes attached to framebuffers:
```
    framebuffer "<fb_name>"
       ...
       attribute [boolean | integer | scalar | string] "<key>" <value>
```
In addition, attribute keyword color respectively vector with 4 respectively 3 values are accepted in the syntax, but ignored in this version.
In order to allow earlier versions of mental ray to parse these files, mental ray version supporting this syntax set the API variable mi_framebuffer_attribute_syntax is set by the kernel. When echoing, the attribute block is protected with $ifdef-ed on that variable.
Experimental: if the boolean attribute "tiled" is set to off, mental ray interprets it as a hint to write out the image in scanline form. Currently, only supported for OpenEXR images. If different values are specified for different framebuffers stored into the same file, tiled takes precedence over scanline. By default, mental ray saves OpenEXR files in tiled format.
Experimental: if an attribute with name staring with "meta:" prefix is used (like attribute string "meta:my_key" "my_value"), the value is passed to the image file headers (i.e. "my_key" "my_value" pair). This is currently only supported for OpenEXR images. Boolean parameters are converted to integer ones.

Bug fixes:

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 rasterizer used in combination with lens shaders, fixed possible black image parts or missing lens shader effect (fix in version 3.9.2.8 is incomplete, extended in version 3.9.2.10).
For unified sampling, added a sanity check on min sampling rate. Fixes possible crash.
For progressive rendering, fixed error estimation code (red channel only was used). This improves sampling distribution and convergence rates for many scenes.
For progressive rendering and iray rendering mode, fixed possible image bit per pixel quality or black images if a color buffer was saved into more than one image file.
Fixed possible black images for the combination of rasterizer and lens shader phenomena.
For MetaSL LLVM backend, fixed "primary output 'result' not supported" errors.
For MetaSL LLVM backend, fixed passing array of unsized arrays.
For MetaSL LLVM backend, fixed possible fatal / segfault for shaders with struct outputs.
For MetaSL LLVM backend, fixed a crash when compiling light profile parameters.
For MetaSL LLVM backend, fixed a crash if compound MetaSL shader output type is not supported by mental ray (like string members).
For MetaSL compiler, fixed possible endless loop when compiling call statement.
For MetaSL compiler, fixed several memory leaks.

Changes in version 3.9.2.7

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 progressive rendering, dithering was no applied. Fixed.
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.
- Improved support for some OEM shaders.
Fixed failure to recompile MetaSL shaders when the set of MetaSL techniques ("metasl technique" string option) is changed from frame to frame.
Fixed loading MetaSL techniques from stings with mi_api_metasl_code() function.
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.

Shader changes:

For userdata shaders, fixed declarations of mib_data_get_shader_* shaders.

Changes in version 3.9.2.6

Feature improvements:

Improved MetaSL compiler reporting (added warnings and notes).

Bug fixes:

On Linux platform, shader libraries are loaded with RTLD_DEEPBIND flag passes to the dlopen(). This prevents possible symbol conflicts defined in both mental ray standalone/application integrating raylib and the shader library.
OpenEXR library integrated into the kernel is compiled in proprietary namespaces starting with "MI". This resolves possible symbol conflicts with further copies of OpenEXR library used in shaders or applications integrating raylib.
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.
Fixed possible excessive tessellation for triangles clipped on camera planes (in combination with rasterizer of fine displacement approximations).
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 thread-safety of some compiler operations. Fixed a potential deadlock.
Fixed several issues in the documentation.

Changes in version 3.9.2.5

Feature improvements:

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.
For unified sampling, fixed per-object min/max sampling rates.
Fixed progressive rendering sampling if min samples is specified to be one sample per pixel.
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 MetaSL shaders, fixed texture lookup with texture filter "none".
Fixed several memory leaks in MetaSL compiler kernel.
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:

For mia_envblur shader, fixed shading is camera instance transformation is not orthonormal.

Changes in version 3.9.2.2

Feature improvements:

Added a limit to prevent tiny Tiff files from being cached for efficiency reasons.
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).

Bug fixes:

Fixed unified sampling used in combination with scanline on option.
For raylib integrations: In the progressive interactive mode, fixed a "progressive cannot resume render" error message.
For MetaSL compiler, fixed internal copying of shaders with compound return types.
Fixed several memory leaks in MetaSL compiler kernel.
For MetaSL compiler, added support for geometry_tangent_u and geometry_tangent_v state variables.

Changes in version 3.9.2.1

Feature improvements:

For iray rendering mode, report total and available memory for each GPU used.
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.
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.
Reduced memory consumption used for error estimation buffers in progressive rendering and iray rendering mode.

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 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 Tiff image files, fixed several issues in reading and caching various Tiff subformats.
Fixed a performance regression when mipmap pyramids were created on demand (i.e. in case of filtered texture lookup to a texture declared unfiltered).
Fixed possible map container crash when a container was used by large number of threads (race condition).
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.
For applications integrating raylib with progressive rendering: fixed possible crash when lights are moved.
For mental ray standalone, fixed possible segmentation fault in case of an fatal error (mi_fatal() call).
Fixed echo of animation in iray mode (-echo iray command line options).
Fixed echo of instances with override material flag.
Fixed several memory leaks in MetaSL compiler.
For MetaSL compiler, fixed compilation error of phenomena encapsulating light shaders.
For MetaSL compiler, fixed possible crash in parser.
For MetaSL C++ backend, fixed possible compilation error for bsdfs.
For MetaSL standalone compiler (ray -mslc), removed obsolete command line options. Fixed compiler message reporting.
For iray rendering mode:
- Fixed a crash when rendering an empty scene.
- Improved driver timeout prevention mechanism in SLI configuration. After a certain number of driver timeout failures the GPU is disabled.
- 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 driver timeout detection failures on CUDA capability 2.0 GPUs.
- Fixed possible crash for baking of textures via GUI attributes introduced in version 3.9.1.24.
- 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.
- Fixed moving lights.

For the imf_disp tool, fixed frame adjustment when the tool is started from a command line.

Shader changes:

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

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 convergence 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.
- "sample 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 if an input parameter of enum type was used.
For MetaSL LLVM backend, fixed 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 NVIDIA ARC, 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 NVIDIA ARC, 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 containing 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 pthread 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 NVIDIA ARC 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 convergence/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.

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: