FlashMandel is "THE" Mandelbrot and Julia fractals rendering program for the
AmigaOS4 platforms :-) It's a very long time project, development started in the
end of '90 when I was sixteen and learned about fractals reading and looking on
specialized newspapers some beautiful pictures. Early develpoment started on an
Amiga 2000 with KS1.3 1MB Ram and SAS/C compiler. FlashMandel was developed by
me, Dino Papararo, in my freetime to learn and improve my skills with C/ASM code
and AmigaOS specific API functions. In a second time joined Edgar Schwan adding
a lot of interesting features like reaction GUI, arexx ports and scripts,
localization, MAPM arbitray precision math, installer script, a dedicated web
page, and many various bug fixes.
So my actual development system is -> Pegasos II G4@1GHz 2GB Ram - Radeon 9250
AmigaOS4.1fe update 2 with latest OS4 SDK and GCC 11.2.0
FlashMandel uses handmade assembler routines for max speed! ..and "SPEED" it's
main feature. With new implemented altivec routines, for altivec version,
renderings are all near realtime! I think it's one of the fastest mandelbrot and
julia sets rendering program ever seen on a single core cpu. Due it's
development for AmigaOS it lacks of multiprocessing features. But once
implemented, it can be adapted for sure to work on more cpu cores.
FlashMandelNG primary features are:
Uses Reaction GUI, Arexx port, Localization, Arbitrary precision math via GMP
Library and don't need any installer, just click 'n run! C and Asm sources are
also included. You can recompile it just installing latest AmigaOS4 SDK and GMP
static LIB included in FalshMandelNG. Copy content in GCC local newlib path just
as like any third party lib add on. FlashMandelNG it's faaast even on QEMU/UAE
emulated Amigas and can render even with direct memory access (TURBO OPTION)
..under x86 emuation it's faster than many others x86 native fractal programs
;-) Saved pictures can be reloaded and further explored at all supported
screen's resolutions. It's really cool and totally free ;-)
In archive we have four binaries:
1) FlashMandelNG_OS4 -> critical routines are written in PPC handmade assembler
to gain max speed on all powerpc cpus families from 603 to G5!
2) FlashMandelNG_OS4_altivec -> critical routines are written in C code but with
full Altivec directives, engine processes 4 pixels per time! Sadly due limited
hardware implementation, precision is limited to float32 datatypes. For Altivec
version ARexx port is disabled, please consider it as an experimental version.
3) FlashMandelNG_OS4_spe critical routines are written in C code but transalted
in SPE native code, it should be faster than FPU version. ** Untested version **
4) FlashMandel_OS4_lite is like standard version but whitout Reaction and Arexx
support, it can have better compatibility under emulated environments like QEMU.
..so what version is the faster one? On a G4 latest FlashMandelNG Altivec
version is really much faster than FPU version because it processes 4 pixels per
time! ..but it's also less accurate due 'float32' single precision calculations
(limited by hardware) vs 'float64' FPU doubles, so for deep zooms you need to
switch sooner to higher precision GMP Library math calculations.
In version 5.1 I have introduced an A1222/Tabor compatiiblity 'wrapper' function
to enable SPE instructions only for core calcs. I'll wait for a new OS4 SDK with
full support for 'A1222' and compiler fixes to enable generation of SPE binaries
for all sources of project and not only two core Mandelbrot and Julia functions.
At this time SPE instructions use scalar capability, but in future I can enable
SPE vector instructions, just like done with Altivec version, in this case with
only one instruction per cycle instead four, but with much better precision due
float64 vs float32 datatypes involved.
About precision, one of main features of FlashMandel, is the ability to use
arbitrary precision math. To archieve this I used GMP library, Edgar did a great
job porting it from linux sources ..and now we have it also on Amiga systems :-)
FlashMandelNG_OS4 is Giftware, If you like it please sent me nice goodies ..An
email or common postcard is ok too! :-) It's possible to include it in magazines
and CDs, if so please let me to have a free copy of paper/media support ;-)
Version 5.1 changes:
Now commandline parameters follows amiga's way rules, for now it's present only
'BENCHMARKMODE ON' option to start FlashMandelNG with an 'FPU', 'Altivec', 'SPE'
and 'Integer Math', speed tests.
Added some checks to validate benchmark results.
Fixed periodicity check and speededup core 'Altivec' routines to use 100% power.
Minor improvementents and bugfixes.
Updated docs.
***
"Bottomless wonders spring from simple rules, which are repeated without end."
cit. Benoit Mandelbrot
Enjoy,
Amiga Rulez!
|
aminet net>