You are reading the change log of a stable version of Hermes. Go to
http://clanlib.org/hermes for development releases.
This file lists the changes to HERMES. Newer versions are listed on top.
-----------------------------------------------------------------------------
Version 1.3.3
- Converted the c_genrc files to macros for greater readability and fixed some bugs
- Added the remaining stretch converters
- Removed the convert.c code - instead it calls the blit.c code
- Improved blit.c converter finder. Uses jump table instead of switch statements
Version 1.3.2
- Maintenance release!
- Fixed alpha blitting.
Version 1.3.1
- Fixed unix build environment.
Version 1.3.0
- Added colorkey and alpha support
- Added blitting support
Version 1.2.6-2 (25/11/1999):
- Fixed Windows compilation problem
Version 1.2.6 (22/11/1999):
- More MMX converters (PII optimised by Jonathan Matthew! Work great on my
k6-2, 20% speedup!)
- Default installation prefix for the RPM files is now /usr not /usr/local
- Bug fixes (thanks to the clanlib people for reports)
- New test program for memory boundary violations
Version 1.2.5 (21/05/1999):
- Included Sam Latinga's changes for static lookup tables.
- Added the first assembler stretching routines ! from 32 rgb 888 to
16 rgb 565 gives a speedup from 30 frames/sec to ~55 frames/sec in a
benchmark over the c routines!
- Patched against a series of patches from Mikko which have major
effects:
- Speed up ALL c routines
- Optimise the case where a scanline is as wide as the pitch..
The whole buffer can be converted at once.. top to bottom.
- When converting from lower bit modes (e.g. 16) to higher bit
modes (e.g.32) some correction is done now
- A fix for the x86 8 bit clearer
- Updated the i8->16 converter (wrote an assembler converter)
- Tuned the generic converters a bit
- Some big endian fixes to the converters
- Casting fixes to reduce warnings with Visual C++ (Gaffer)
- Allowed the index parameter to the clear routine to be used as an alpha
parameter (optionally)
- Updated the API specification with the index change and spell-checked it :)
- Sped up the 32rgb888 to 24rgb888 C converter and the C i8 to 16rgb565
converter.
- Included Gaffer's windows Visual C++ project file and updated the README
file. NASM is now required for Visual C.
- Upgraded libtool to 1.3.
- Fixed the assembler stretching routines. They don't use globals any more
and should be thread-safe now.
- Added Hermes_PaletteInvalidateCache.. sorry about that one.. it was
missing in the last release.
- There's another public include file now, H_Config.h that concerns
information which unfortunately has to be exported from Hermes. (only
one #define so far anyway).
- Sped up the i8 to 24 C converter (write 4 pixels at a time). we were
a bit lazy about C converters to 24 bit admittedly.. well, we've got
the assembler stuff anyway.. :)
- Changes changes changes.. Gaffer fixed up the visual c++ stuff a bit
more and Types.h is now *deprecated*.
- C++ styles comments (//) removed and replaced. Doh, seems some C
compilers are not up to the latest standard (yes, // comments are in
the latest C standard!)
- extern "C" in front of all declarations if you compile with a C++
compiler..
Version 1.2.4 (22/02/1999):
- Included debugging macros from the development version after complaints
from windows and dos programmers who aren't used to seeing debugging
output ;)
- 100000 thanks to Mikko for the intel2gas converter, NASM is NOT
required anymore under Linux (or djgpp) for assembler support !!!
- Palette lookup generation will respect alpha values now (there you are
Gaffer..)
- Watcom problems with stack calling fixed.
- Added some documentation for the public interface into the source code.
Yes, you don't have to look it up in the API every time anymore now :)
- Factory converters will have the alpha mask set properly now. Beware as
RGBA888 now has the mask 0xff000000,0xff0000,0xff00,0xff, where the last
one was 0 previously.
- In addition, converters to RGBA and BGRA will now set the alpha bits to
1 instead of 0. Note that there are no converters at the moment to
convert to the formats ARGB888 and ABGR888. Only RGB888 and BGR888 are
available. (Everybody seems to use RGBA anyway..)
- Fixed a nasty NASTY MMX bug that occurs in Hermes 1.2.0-1.2.3.
- Compiles under FreeBSD!
Version 1.2.3 (18/01/1999):
- Hermes development has forked into a stable branch (this), 1.2 and a
development branch 1.3. This is because the features in 1.3 (YUV
colour space, tiled rendering etc.) would have broken the interfaces
of some libraries.
- Switched the whole thing to libtool. Please bear with us if Hermes does
not compile on your non-Linux UNIX system now. We're working on it
(automake's and libtool's fault! Cross-platform? Doesn't work on IRIX
or SunOS or Solaris or FreeBSD !!!)
- Quite a few converters integrated from the development version. All
formats are covered now (apart from stretching). Only optimisation left.
- Added a Makefile for Visual C. Untested.
Version 1.2.2 (30/12/1998):
- install / uninstall added. have fun.
- Hermes.spec added, create your own RPMs if you want.
- More stretch converters, from Mikko
- Switched to stack passing from register passing. It creates more
problems than good and only one pointer is being passed anyway ...
- Fixed a few nasty bugs with 8 bit converters where two pixels wouldn't
be drawn at the end of scanlines
- Finished the API documentation for now.
Version 1.2.1 (7/12/1998):
- Implemented a few stretching routines to show the way (All straight
copy routines except 24 bit are stretching, as is 32to16, 32to8 and
indexed8 to 16)
- Rewrote converter assignment in Factory.c (look into FactConv.c, that
now holds an array of all converters, with priority).
The switch statement from hell is gone. R.I.P.
- Watcom and DJGPP makefile integrated (see CONTRIB)
- Assembler routines patched to use shorter opcodes (see CONTRIB)
- Bugfixes in HeadX86.h (see CONTRIB :)
- Bugfix in c_clr.c, 32 bit clearer works OK now (thanks zoon)
- Bugfix in HermConf.h (djgpp..)
- Added --disable-x86asm to build the library without assembler support
on all unix systems, even if nasm is detected (good for testing and
benchmarking)
- There will now be an attempt to autodetect the system type when
configuring for assembler support, using the uname program. It should
work correctly for Linux and FreeBSD. If it doesn't, you'll be prompted
instead.
- Hermes will be autoconf'd for C instead of C++ from now on. You have to
override this manually or use #extern "C" in your C++ code that includes
Hermes. (unix only)
- Some preliminary dithering routines, from code by the rasterman.. Ignore
them, they're not intended for use.
- Started API documentation in SGML.. Look into docs/api for a text file
and html version.
- Added FAQ
Version 1.2.0 (13/11/1998):
- Friday the 13th release of a COMPLETE REWRITE, you may estimate the
number of bugs yourselves :)
- Support for optimisated routines for different processors
- Better support for stretching
- Assembler #ifdefs pinpointed and localised in one .c file (apart from
the headers that declare the routines)
- Support for both converters that have a main loop that calls a special
loop for every scanline and all-in-one converters that will convert a
whole rectangular block
- A factory method to configure Hermes with the right routines at startup,
rather than on the fly.
- Cached palette lookup tables, 'nuff said :)
- Frequent-Conversion-Friendly system that will retrieve converters for
popular colour formats F.A.S.T.
- The dreadful #define NO_ASSEMBLER has been replaced by one that means
the opposite: #define X86_ASSEMBLER. Omitting it will not produce errors
anymore but rather compile correct C routines
- Some bugs from previous versions eliminated
- Not tested to the same extent as the previous version. Handle with care
and be prepared for a quick follow-up release.
Version 1.0.4 (never released):
- Added Mutable.c that implements a list of per-instance structures
- Most functions return chars now to indicate whether they failed
- HermesInit returns a handle now
- More checks whether things get initialised properly
- Added assembler routines for conversion from 32 bit to: 32 bit BGR,
32 bit RGBA, 32 bit BGRA (from ptc 0.72)
Version 1.0.3 (20/10/1998):
- 24 bit Generic to * routines added. Indeed, ANY 32, 24, 16 bit direct
source format will now be converted to any 32,24,16,8 bit destination
format (by a generic routine in the worst case)
- Lots of big-endian fixes (this time they are tested, on an SGI O2).
Still quite a bit of big-endian testing left though..
- Clear support added ! (C routines so far)
- W00p.. Assembler clear routines
- W000000p.. MMX clear routines ! :) (MMX actually pays for memory
clearing. All the MMX routines, with no exceptions, are faster than
x86 or C routines at surface clearing!)
- Plenty of stuff planned for the next version, don't even dream of
a frozen API :)
Version 1.0.2 (01/10/1998):
- Muhmuh renamed to Muhmu (DOH :)
- 16 bit to * assembler routines added (adapted from PTC 0.72)
- 'Multiple users' support added. You can initialise Hermes more than once
now, it will keep a reference count. When there are no users left, then
it will be destroyed again.
Version 1.0.1 (17/9/1998):
- Different C format conversion routines separated into different files.
(c_16.c c_32.c etc)
- Little/big endian distinction handled by macros in a more efficient way.
- TODO.conversion created to keep track of the status of conversion
routines.
- Destructor bug fixed
- PLENTY of new C routines added and old routines optimised
- All char* replaced with char8* (fixes a colour overflow bug)
- Muhmuh colour format added
- Generic To Generic routines added
- New assembler routines
- Bugfix, MMX routines work now
- 32 to 16 MMX routine temporarily removed
Version 1.0.0 (8/9/1998):
- Initial release to the public
