diff options
author | Tom Schouten <doelie@users.sourceforge.net> | 2003-01-21 10:27:33 +0000 |
---|---|---|
committer | Tom Schouten <doelie@users.sourceforge.net> | 2003-01-21 10:27:33 +0000 |
commit | 9b8745d5250c9d0b60c9aa5a77f58a3fcddf1076 (patch) | |
tree | 8372b6a414a7124cec57efc9c80845e2bc1b157d /scaf/README.scaf |
This commit was generated by cvs2svn to compensate for changes in r352,svn2git-root
which included commits to RCS files with non-trunk default branches.
svn path=/trunk/externals/pdp/; revision=353
Diffstat (limited to 'scaf/README.scaf')
-rw-r--r-- | scaf/README.scaf | 98 |
1 files changed, 98 insertions, 0 deletions
diff --git a/scaf/README.scaf b/scaf/README.scaf new file mode 100644 index 0000000..0035899 --- /dev/null +++ b/scaf/README.scaf @@ -0,0 +1,98 @@ +SCAF - simple cellular automaton forth + +scaf is a virtual machine / forth environment for binary arithmic +tailored to 2D 1 cell neighbourhood cellular automata. + +scaf is a compiled language. programs run inside a "feeder" +(sort of operating system if you want) + +the feeder is responsable for loading/storing CA cells +from/to memory. data in memory is organized as a scanline +encoded toroidial bitplane (lsb = left). to simplify the feeder +and the stack machine, the top left corner of the rectangular grid +of pixels will shift down every processing step. this enables +to keep a cell neighbourhood in a couple of registers. + +the stack machine has the following architecture: +CA stack: (%esi), TOS: %mm0 (32x2 cells. lsb = top left) +CA horizon: (%edi) (64x4 cells. (%edi) = top row. lsb = left) + +scratch register: %mm1, %mm2 +bitmask register: %mm3 = 0xffffffffffffffff + +4 bit counter: %mm4-%mm7 + +the stack size / organization is not known to the stack machine. +it can be thought of as operating on a 3x3 cell neightbourhood. +the only purpose of the forth program is to determine the CA local update rule. + +the machine is supposed to be very minimal. no looping control. +no adressing modes. no conditional code. so recursion is not allowed +(no way to stop it) there are 9 words to load the cell neigbourhood +on the stack. the rest is just logic and stack manips. + +the counter can be used for counting neighbourhood cells, like in the +game of life. the zero test and sign bit can be used for comparisons. +there are kernel words for loading constants into the counter register, +and for communication between stack and register. + +the horizon is limited to 3x3, however it can be easily extended to +32x3. extending it further than that would require a redesign of the +forth + feeder. + + +HOW TO CREATE NEW CA RULES + +edit scaf/modules/carules.scaf or create your own source lib and add +the name to the scaf/modules/Makefile. type make in scaf/modules +to compile. if you get error messages from the assembler saying things +like + + Error: no such instruction: `xxx' + + or + Error: invalid character '_' in mnemonic + +this means there are undefined words in your source file. since not +all characters are allowed in an asm file, the offending characters are +converted to _SOMETHINGELSE_ + +if you are stuck somewhere, just look at the output of scaf.pl on +your .scaf file to determine where the problem is. + +words that can be accessed from inside pdp_ca have to start with the +prefix rule_ and have to leave a single item on the data stack (the return +value) other rules can have all the stack effect you want, but for safety +reasons they can't be accessed from within pdp_ca. + + + +FORTH SYSTEM CODE + +the forth system is made up of the following files: + +kernel.scaf: a collection of forth kernel words +scafmacro.s: a set of asm macro definitions used in kernel.scaf +optim.rules: some substitution rules to eliminate redundant + stack manipulations + +scaf.pl: the compiler + +scaf.pl is run like this: + +scaf.pl -Isystemdir source.scaf + +if the -I switch is left out, the current directory is searched +for the system files. the compiler produces an assembler source +that includes scafmacro.s on standard output. + +the code it produces is relatively fast. it only uses and/or/xor +and shift mmx instructions. it's not optimal use of silicon but +it's pretty fast given what's possible. the feeder routine could +be improved though. + +porting to another platform would require a rewrite of scafmacro.s +the rest can be reused. if the target machine has 64 bit registers +(or if you can emulate this one using more registers) porting is +relatively easy. for larger registers a small change needs to +be made to the feeder routine in pdp_ca.c |