--- ray/doc/notes/ReleaseNotes	2023/11/27 22:04:45	1.433
+++ ray/doc/notes/ReleaseNotes	2025/05/21 20:42:48	1.489
@@ -2634,5 +2634,189 @@ that made reds into blues and vice versa, which has be
 Added i/o for spectral pictures to rmtxop, the first program that can
 ingest them.  (Only rtrace, rcontrib, and rfluxmtx can render them.)
 New symbolic rmtxop -c options, such as RGB, XYZ, S (for scotopic)
-and M (for melanopic) make color conversions simpler.
-Also added spectral picture support to rcrop and radcompare.
+and M (for melanopic) make color conversions simpler.  Alternatively,
+a reference matrix or picture may be given and the color space of
+that file will be used.  The -C option takes either a symbolic
+color space or reference file and is "sticky," so it applies to
+all subsequent input files that do not have their own -c right ahead
+of them.
+
+Added spectral picture support to rcrop and radcompare.
+
+Added new utility, rcomb, which is something like a combination
+of rmtxop and pcomb, it works with all the inputs and
+outputs that rmtxop handles, but allows for general expressions
+like those in pcomb for putting matrices together, working on
+an element at a time.  (No resizing of matrices or accessing
+neighboring elements, as permitted in pcomb.)  It saves memory
+over rmtxop by operating on a row at a time, so can handle
+input and output matrices with any number of rows.  It does
+provide for a single matrix-multiplication operation at the end,
+which only requires memory for that final matrix.
+
+Added support for filtering hyperspectral images to pfilt.
+
+Added support for 1- and 3-component float matrix i/o to
+pcomb to allow for better interoperability.
+
+Added support for -co+ option to rtpict, so it can directly
+produce hyperspectral images.
+
+Added new "specdata" and "specpict" primitives to support
+interpolated spectral data files and hyperspectral images
+as patterns.
+
+Added support for multispectal sampling in rsensor, which will
+put out as many color channels as specified in any -cs option
+(default is 3 for RGB).
+
+Added ability of mgf2rad to convert spectral data at least
+for light, glow, plastic, and metal primitive types.  Need
+to use new '-s' option to get new output.
+
+Added handling of spectral images to rcode2bmp script.
+
+Minor change to rcollate to allow explicit byte records to
+be any specified length without checks against header input.
+
+Added new hemisphere type to rfluxmtx, "h=cie" for CIE sky
+scanner positions.  These are 145 patches with identical
+row counts to Tregenza, but starting from a different
+azimuth and reversing row directions at adjacent altitudes.
+
+Added optimization for long argument lists and arrays in
+the calcomp routines used for rendering and in rcalc, etc.
+
+Added error estimate spreading to ambient super-sampling
+algorithm (-as) to improve performance especially in
+scenes with sharp cut-offs.
+
+Added -i option to gendaymtx to match -i option of gendaylit.
+Thanks to Yongqing for initial implementation.
+
+Changed rcrop so negative #rows or #cols counts back from end.
+
+Added rcomb -n option for multi-processing of very large matrices.
+
+Made it so rsplit and rlam can handle any number of input files
+up to active open file descriptor limit.
+
+Taoning Wang added genssky tool for generating spectral sky
+descriptions.
+
+Taoning added gensdaymtx utility based on gendaymtx, and
+added epw2wea -a option to include required information.
+
+Fixed bug introduced in Auguest 2022 (5.4 release) in distant
+light source sampling, which caused aiming failures.  (Thanks
+to Mike DiPompeo of LightStanza for reporting the issue.)
+
+Added rxpict rendering tool based on C++ RpictSimulManager class.
+Multi-processing and spectral output are supported.
+
+Wrote rxpiece tool that will render pictures using tiling and
+can handle different output formats and hyperspectral pictures.
+Only runs on a single machine, however -- no NFS locking on
+shared volumes.  Mainly a demonstration tool for RpictSimulManager
+C++ class, but may have its uses.
+
+Fixed a bug in BSDF sampling of sources pointed out by
+David Geisler-Moroder, which tended to overestimate specular
+contributions for some BSDFs.
+
+Added crude ability to read and convert spectra to RGB in ximage,
+ra_bmp, ra_ps, and ra_rgbe programs.  Also updated fast tone-mapping
+library to handle spectral HSR inputs at the same time.  HSR
+input handling was added in a similar fashion to pcond and pvalue
+for convenience.
+
+Added handling of hyperspectral images to normtiff, a handy program
+for quickly tone-mapping Radiance RGBE, XYZE, and now hyperspectral
+pictures, as well as IEEE float or LogLuv TIFFs.
+
+Added handling of hyperspectral pictures to pextrem, though
+results are still reported in RGB.
+
+Added accurate conversion of hyperspectral pictures to ra_xyze.
+
+Added hyperspectral picture conversion to ra_tiff, with accurate
+color when targeting 16-bit/comp., LogLuv, and float utput,
+and the approximate method for 24-bit RGB.
+
+Taoning added genssky -L option to specify global and diffuse
+illuminance values for output calibration.
+
+Created C++ class for rcontrib and example tool rxcontrib, which
+is an optional compile in the src/rt directory.  The output
+model is more general in some respects and the operations more
+efficient, but standard output is not supported, nor is ASCII
+results data.
+
+Added strnstr.c compatibility module in ray/src/common, since
+it isn't present on Linux derivatives.
+
+Added WGMDfunc material type with programmable roughness and
+separate modifier paths for the different components.
+
+Added gensurf -i option at the request of Peter Apian-Bennewitz
+to reverse surface normals, mostly for data input, when it is
+a bit more difficult to adjust.
+
+Added rad -N option to rholo call for multiprocessing when
+mkillum and/or mkpmap are involved in octree prep.
+
+Added ability of gendaymtx to read EPW as well as WEA files.
+When EPW input is provided, the 3-hour average dew point is
+used to compute a more accurate precipitable water content
+value.  It currently does not use these data directly from
+the EPW input, relying instead on a dew-point-to-W calculation.
+This follows the request and recommendations of David G-M.
+
+Removed limitation in rfluxmtx, which was disallowing distant
+and local receivers. Found this to be a perfectly valid approach
+for some annual daylighting scenarios.
+
+Wrote new pvsum tool to handle cases dctimestep can't, like
+spectral input pictures.  It also adds a -N option for multi-
+processing on Unix machines, which utilizes memory-mapped
+inputs for efficient i/o sharing.
+
+Made it so genBSDF can use spectral rendering for a
+more accurate photometric result.
+
+Changed default matrix representation to float rather than
+double, which uses less RAM and supports faster operations
+in general.  Build with compiler flag:
+	-DDTnativeDTrmx_native=DTdouble
+to get back higher precision representation.
+
+Taoning Wang updated genssky and gensdaymtx to provide illuminance
+calibration and accept EPW files directly as input.
+
+Added -e expr and -f file.cal options to all rendering tools, not
+just rcontrib/rxcontrib.  Feature request from Peter A-B.
+
+Fixed slight bias in indirect calculation caused by ambient
+collision detection, pointed out by Jon Sargent.  New code
+keeps track of rejected samples and sells them back to
+later calls to get something closer to a true Poisson
+sampling method.
+
+Altered rvu "origin" command to take shift amounts forward,
+right, and up rather than taking an explicit origin, which
+can be done easily enough with the "view" command.  Peter A-B
+requested (sort of).
+
+Added genglaze tool created by Taoning Wang to generate
+spectral glazing materials based on CGDB data and models.
+
+Fixed bug that caused inf and nan errors with -cw setting
+where longer wavelength was substantially less than limits
+of specified spectra.
+
+Created bsdfpeaks utility to identify and report full-width,
+half-maximum sizes of peaks in SIR and XML inputs.
+
+Changed default super-sampling in bsdf2ttree from 256 to 64,
+since there doesn't seem to be much benefit to the higher
+sampling rate for most BSDFs.