Imath is maintained by the OpenEXR project, a part of the Academy Software Foundation. The library were originally developed at Industrial Light & Magic and first released as open source in 2003.
Read the origin story of OpenEXR and Imath, and the
float type, in particular, on the ASWF Blog.
Imath is Version 3 because it was previously distributed as a component of OpenEXR v1 and v2.
OpenEXR and Imath is included in the VFX Reference Platform.
New Features in 3.1¶
The 3.1 release of Imath introduces optimized half-to-float and float-to-half conversion using the F16C SSE instruction set extension, if available. These single-instruction conversions offer a 5-10x speedup for float-to-half and 2x speedup for half-to-float over Imath/half’s traditional table-based conversion (timings depend on the data).
In the absence of the F16C instruction set, the lookup-table-based conversion from half to float is still the default, but Imath 3.1 also introduces an optimized bit-shift conversion algorithm as a compile-time option that does not require lookup tables, for architectures where memory is limited. The float-to-half conversion also no longer requires an exponent lookup table, further reducing memory requirements.
These new conversions generate the same values as the traditional methods, which ensures backwards compatibility. See Install for more installation and configuration options.
half.h can now be included in pure C code for a definition
of the type and for conversions between half and float.
OpenEXR/Imath 2.x to 3.x Porting Guide¶
See the OpenEXR/Imath 2.x to 3.x Porting Guide for help in restructing old code to work with recent releases of OpenEXR and Imath.
The ILM Imath library and the
half data type were originally
designed and implemented at Industrial Light & Magic by Florian Kainz,
Wojciech Jarosz, Rod Bogart, and others. Drew Hess packaged and
adapted ILM’s internal source code for public release.
For a complete list of contributors see CONTRIBUTORS.md.