diff options
| -rw-r--r-- | cart.txt | 92 | ||||
| -rw-r--r-- | messages.pl | 17 | ||||
| -rw-r--r-- | textdecomp.s | 147 |
3 files changed, 155 insertions, 101 deletions
@@ -4,15 +4,12 @@ What's needed to get taipan onto a cart: joey_z is willing to manufacture carts like this: +---------------------------------------------------------------------------+ -| Type 13: XEGS 64 KB cartridge (banks 0-7) | +| Type 12: XEGS 32 KB cartridge | +---------------------------------------------------------------------------+ - One of the two variants of the 64 KB XEGS cartridge, that's built on either - a C100649 board with the W1 solder point not connected, or a C026449 board - with pin 9 of the 74LS374 chip unconnected. This bank-switched cartridge occupies 16 KB of address space between $8000 - and $BFFF. The cartridge memory is divided into 8 banks, 8 KB each. - Bank 7 (the last one) is always mapped to $A000-$BFFF. Three lowest bits of + and $BFFF. The cartridge memory is divided into 4 banks, 8 KB each. + Bank 3 (the last one) is always mapped to $A000-$BFFF. Two lowest bits of a byte written to $D500-$D5FF select the bank mapped to $8000-$9FFF. The initially selected bank is random, although it seems that 0 gets chosen the most often. Atari800 always selects bank 0 initially. @@ -31,27 +28,28 @@ the 1200XL. ...so: -Bank 7 has the startup code, uncompressed title screen, title code, and -code to copy stuff from the other banks to RAM. The main portion of the -code gets split up into bank-sized chunks... but the last bank of code -doesn't need to be copied to RAM: I use a custom linker script to have -cl65 org the RODATA segment there, plus a new segment called HIGHCODE, -which is executable code that will run directly from the cartridge. This -bank stays selected while the game is running, so it also has the font -in it. - -There are currently 3 banks of code (numbers 0, 1, 2) that need to be -copied to RAM. The bank with RODATA and HIGHCODE is bank 3. Banks 4, 5, -6 are unused. As I optimize the code, bank 2 will contain less and less -code, and hopefully will disappear one day (at which point, I only need -a 32K cart instead of a 64K one). - -Code in bank 7 will copy all the chunks to correct place in RAM... and +Bank 3 has the startup code, uncompressed title screen, title code, +and code to copy stuff from the banks to RAM, and the tail-end of the +code that needs to be copied. The main portion of the code gets split up +into bank-sized chunks... but the last bank of code doesn't need to be +copied to RAM: I use a custom linker script to have cl65 org the RODATA +segment there, plus a new segment called HIGHCODE, which is executable +code that will run directly from the cartridge. This bank stays selected +while the game is running, so it also has the font in it. + +There are currently 2 full banks of code (numbers 0, 1) that need to be +copied to RAM. The bank with RODATA and HIGHCODE is bank 2. + +Code in bank 3 will copy all the chunks to correct place in RAM... and I don't need to leave room for DOS or anything else, so the code can be ORGed at $0400 (romable_taimain.raw target in the Makefile does this). -Currently, romable_taimain.raw is 18739 bytes. $0400 + 18739 means the -code ends at $4d33. The BSS is right after that, and is less than a +romable_taimain.raw is the code that's org'ed to $400, that gets copied +to RAM. rodata.8000 is the code/data that is used directly from ROM, +in bank 2. + +Currently, romable_taimain.raw is 17251 bytes. $0400 + 18739 means the +code ends at $4763. The BSS is right after that, and is less than a page. The OS will place the GR.0 display list at $7c20, and the stack will grow down from there to $7a20 (except it never grows that much). @@ -65,26 +63,20 @@ Amusingly, the Taipan cart will work on a 32K 800. Not a 16K Atari though. any stock 24K Ataris anyway, they'd be 8K or 16K machines with a 3rd-party RAM upgrade, and exceedingly rare these days). -bank 7: fixed bank +bank 3: fixed bank $a000-$bxxx - title screen data, dl, menu code, memory size checker, code to copy romable_taimain to RAM. - currently 1441 bytes free in this bank. +$bxxx-$bfff - tail end of romable_taimain.raw (around 1400 bytes) $bffa-$bfff - cart trailer banks 0, 1: full banks of romable_taimain code $8000-$9dff - 31 pages (7936 bytes, 7.75K) of code $9f00-$9fff - unused, filled with $ff -bank 2: last (partial) bank of romable_taimain -$8000-$9f00 - up to 31 pages (7936 bytes, 7.75K) of code. - currently only 2809 bytes (11 pages) are used, - and the number's getting smaller all the time. -$9f00-$9fff - unused, filled with $ff - -bank 3: RODATA and HIGHCODE segments. +bank 2: RODATA and HIGHCODE segments. This bank stays enabled after copying is done. - currently all but 7 bytes are used. + currently all but 200-odd bytes are used. $8000-$9bff - up to 7K of code/data (28 pages) $9c00-$9fff - font (1K). The font has a nonzero byte at $9ffc, so the OS doesn't think there's a cartridge here @@ -94,15 +86,12 @@ Unused areas are filled with $ff (this is the default state for both flash and EPROM). For the banks that map at $8000, this includes the cart trailer area. A non-zero byte (our $ff) at $9ffc tells the OS that a cart isn't inserted in the right slot, so it won't try to initialize/run -our cart as a right cart. Only bank 7 (that maps as a left cart) has a +our cart as a right cart. Only bank 3 (that maps as a left cart) has a valid cart trailer... according to cart.txt, every once in a while, bank -7 might come up selected in the $8000 bank at power on. This shouldn't +3 might come up selected in the $8000 bank at power on. This shouldn't matter: it'll be in both bank areas, and if the OS tries to init it as a right cart, the init/run addresses will point to the left cart area. -banks 4, 5, 6 are unused (24K total). Possibly the manual goes here, -if I write one. - -- Changes the game needed for a cart version: Not many. @@ -117,20 +106,20 @@ all of the game's asm code, and a small amount of the C code, goes in a new segment called HIGHCODE (see cartbank3.cfg). checkmem.s isn't needed any longer... though there is a new memory checker -(in bank 7) that says "32K required" if someone tries it on a 16K machine. +(in bank 3) that says "32K required" if someone tries it on a 16K machine. -Exiting the game (N at "Play again?" prompt) returns to the title screen, -since there's no DOS to exit to. This might be slightly useful: you -might decide to change the colors or disable sound for your next game. +Exiting the game (N at "Play again?" prompt) reboots since there's no +DOS to exit to. You end up at the title screen again. Pressing the Reset key does a coldstart (reboot) in the cart version. You end up at the title screen again. -The title decompression is gone: it just displays the title screen DL -and data straight from ROM. The menu help text and the tail end of the -display list are in RAM though, so the menu can change them. Moving the -end of the DL to RAM means one extra black (border-colored) scanline -appears due to the DL jump instruction. +The title decompression is gone: it just displays the title screen DL and +data straight from ROM. The menu help text and the tail end of the display +list are in RAM though, so the menu can change them. Moving the end of +the DL to RAM means one extra black (border-colored) scanline appears +due to the DL jump instruction. This is the only visible difference +between the cart and xex builds. The font is located in ROM, on a 1K boundary, so CHBAS can point to it, rather than $2000 like the xex version. @@ -138,9 +127,6 @@ to it, rather than $2000 like the xex version. num_buf and firm are located in the BSS rather than page 6 as they are in the .xex version. -Since I have 3 empty banks... Why not include a manual on the cart, -with pseudo-hypertext UI? - -- What would be *really* slick: figure out a way to split the code up @@ -156,7 +142,9 @@ in asm? Probably. Do I want to? Not really. A 5200 version should be possible. At the moment, the .xex version is just under 32K (excluding the memory-checker segment). The 5200 would need its own conio, since cc65's conio for 5200 uses a 20x24 GR.1 screen -and doesn't support input at all. +and doesn't support input at all. It would also needs its own bignum +library, since the bigfloat one uses the OS's floating point ROM (which +doesn't exist on the 5200). Probably will do a bigint48 bignum lib. The 5200 cart window is 32K, so no bankswitching would be required. There's only 16K of RAM, but that'll be plenty. diff --git a/messages.pl b/messages.pl index 0d68f4d..1120a8b 100644 --- a/messages.pl +++ b/messages.pl @@ -2,7 +2,11 @@ # compresses messages for taipan.c. # messages are listed at the end of this file after __END__ marker. +# Run with no arguments to encode all messages, in which case the # output of this script should be redirected to messages.c. +# With an argument, encoding is not done: instead, the strings +# to be encoded are dumped to stdout, *after* dictionary +# substitution is done. # make dictionary from textdecomp.s comments open my $t, "<textdecomp.s" or die $!; @@ -35,7 +39,7 @@ while(<DATA>) { s/^\w+\s+//; my $orig = $_; - #warn "msg: $_\n"; + print " input: $_\n" if @ARGV; s/"//g; s/\\r//g; s/\\n/\n/g; @@ -49,9 +53,12 @@ while(<DATA>) { } } - my $w = $_; - $w =~ s/\n/\\n/g; - #warn "got: \"$w\"\n"; + if(@ARGV) { + my $w = $_; + $w =~ s/\n/\\n/g; + print "output: \"$w\"\n\n"; + next; + } open my $out, ">msg.out" or die $!; print $out $_; close $out; @@ -69,6 +76,8 @@ while(<DATA>) { print "\n};" . ($dict_used ? " // dictionary used" : "") . "\n\n"; } +exit 0 if @ARGV; + print "// messages: $msgcount\n"; print "// total input size: $total_in\n"; print "// total output size: $total_out\n"; diff --git a/textdecomp.s b/textdecomp.s index 617ba2f..0a0c684 100644 --- a/textdecomp.s +++ b/textdecomp.s @@ -1,25 +1,64 @@ -; text decompressor for taipan. -; text is packed 6 bits per character. see textcomp.c -; for details. +; Text decompressor for Taipan. + +; extern void __fastcall__ print_msg(const char *msg); + +; Text is packed into one snac per character. + +; A snac is 6 bits, somewhere between a nybble and a byte. It could +; also stand for "Six Numeral ASCII-like Code" :) + +; See textcomp.c for details of encoded format. .include "atari.inc" .export _print_msg .import _cputc srcptr = FR1 - outbyte = FR0 ; decoded 6-bit byte + outsnac = FR0 ; decoded snac (6-bit byte) bitcount = FR0+1 ; counts 8..1, current bit in inbyte - inbyte = FR0+2 + inbyte = FR0+2 ; current input byte ysave = FR0+3 - dict_escape = FR0+4 + dict_escape = FR0+4 ; true if last character was a Z .rodata -; one or two letter words are not worth listing here. 3 is only good -; if it's used pretty often. -; entry 0 is a dummy! The encoder gets confused by "Z\0". This may get fixed. -; dictionary size cannot exceed 255 bytes. -; the quoted stuff in comments is read by messages.pl, it needs to be exact. + +; The dictionary itself. Each entry is a snac-encoded string. One or two +; letter words are not worth listing here: they encode to 2 bytes each, +; plus the dictionary escape code is 2 bytes (snacs actually) per use. 3 +; is only good if it's used pretty often. + +; In messages.c, dict01 to dict26 will show up as Za thru Zz, and dict27 +; and up are ZA, ZB, etc. In theory, a dict entry could reference another +; dict entry (the decoder can handle it), but in practice it's not real +; useful to do. + +; Entry 0 is a dummy! The encoder gets confused by "Z\0". This may +; get fixed. + +; There can be be up 63 entries in the dictionary (64, counting the +; dummy entry 0), since a 6-bit snac is used as the index. + +; Dictionary size cannot exceed 255 bytes. Actually the last entry +; can extend past 255 bytes, so long as it *starts* within 255 bytes +; of dict00. Break this rule and you get a range error when you build. + +; The quoted stuff in comments is read by messages.pl, it needs to be +; the exact un-encoded form of the snac string. Anything after the quotes +; (e.g. number of occuurences) is ignored. The order here isn't important, +; messages.pl will apply them in order by length (longest first). + +; To get the bytes to use for a particular message: +; echo -n "message here" | ./textcomp 2>/dev/null|perl -ple 's/0x/\$/g; s/ /, /g' + +; TODO: no way to use \n in these (which affects dict33), fix. + +; TODO: if a message used in the game is exactly the same as a dict entry, +; figure out a way for the game to use the dict entry in-place, instead +; of a string consisting only of a dictionary lookup. Perl script can +; generate an asm file that gets included here? _M_taipan = dict23, and +; in messages.c it's an extern. + dict00: dict01: .byte $98, $9d, $73, $54, $53, $80 ; "Li Yuen", 4 occurrences dict02: .byte $7c, $c1, $05, $4b, $57, $12, $3d, $42, $05, $48, $00 ; "Elder Brother", 3 @@ -52,6 +91,14 @@ dict28: .byte $d5, $70, $4e, $50, $00 ; " want" dict29: .byte $5c, $f4, $94, $20, $93, $85, $4d, $30, $00 ; "worthiness" dict30: .byte $d4, $d5, $43, $20, $00 ; " much" dict31: .byte $10, $91, $86, $15, $21, $4e, $0c, $50, $00 ; "difference" +dict32: .byte $74, $f3, $50, $48, $11, $0f, $48, $00 ; "Comprador" +dict33: .byte $f1, $3d, $6c, $15, $03, $d2, $50, $00 ; "'s Report" +dict34: .byte $6d, $91, $78, $d5, $71, $7c, $30, $cd, $40 ; "Aye, we'll " +dict35: .byte $08, $f0, $52, $10, $00 ; "board" +dict36: .byte $40, $94, $81, $50, $50, $00 ; "pirate" +dict37: .byte $d4, $e3, $c0 ; " no" +dict38: .byte $d5, $72, $53, $20, $00 ; " wish" +dict39: .byte $10, $50, $94, $00 ; "debt" dict_offsets: .byte dict00 - dict00 @@ -86,6 +133,14 @@ dict_offsets: .byte dict29 - dict00 .byte dict30 - dict00 .byte dict31 - dict00 + .byte dict32 - dict00 + .byte dict33 - dict00 + .byte dict34 - dict00 + .byte dict35 - dict00 + .byte dict36 - dict00 + .byte dict37 - dict00 + .byte dict38 - dict00 + .byte dict39 - dict00 ; rough estimate of how many bytes are saved by the dictionary ; stuff: the dictionary + extra decoder stuff costs 221 bytes (vs. @@ -94,13 +149,13 @@ dict_offsets: ; with only dict00 - dict23, we'll save around 173 bytes. ; actually it works out to 179 bytes, but the estimate was close. ; we've reached the point of diminishing returns: dict00 - dict31 only -; saves 200 bytes. +; saves 199 bytes. dictsize = * - dict00 .out .sprintf("dictionary plus dict_offsets is %d bytes", dictsize) .rodata -table: ; outbyte values 53..63 +table: ; outsnac values 53..63 .byte ' ', '!', '%', ',', '.', '?', ':', 39, 40, 41, $9b tablesize = * - table @@ -110,58 +165,57 @@ table: ; outbyte values 53..63 .code .endif -; extern void __fastcall__ print_msg(const char *msg); _print_msg: sta srcptr stx srcptr+1 lda #0 sta dict_escape - sta outbyte + sta outsnac ldy #$ff ; since we increment it first thing... - ldx #6 ; counts 6..1, current bit in outbyte + ldx #6 ; counts 6..1, current bit in outsnac @nextbyte: iny lda #8 - sta bitcount + sta bitcount ; counts 8..1, current bit in inbyte lda (srcptr),y sta inbyte @bitloop: - asl inbyte - rol outbyte + asl inbyte ; get next bit from inbyte... + rol outsnac ; ...into outsnac dex - beq @decode ; got 6 bits - dec bitcount - bne @bitloop - beq @nextbyte + beq @decode ; got 6 bits, decode into ascii + dec bitcount ; more bits in this byte? + bne @bitloop ; get rest of bits in this byte... + beq @nextbyte ; ...else next byte @decode: - lda outbyte - bne @notend - rts ; 0 = end of message + lda outsnac + bne @notend ; are we done? + rts ; 0 = end of message @notend: - ldx dict_escape ; was last character a Z? + ldx dict_escape ; was previous character a Z? beq @normalchar - jsr dict_lookup - jmp @noprint + jsr dict_lookup ; if so, do a dictionary lookup... + jmp @noprint ; ...and pick back up at next byte -@normalchar: +@normalchar: ; else it's a normal character cmp #27 bcs @notlower - adc #'a'-1 ; 1-26 are a-z + adc #'a'-1 ; 1-26 are a-z bne @printit @notlower: cmp #52 bne @notdict inc dict_escape ; Z means next 6 bits are dictionary ID - bne @noprint + bne @noprint ; don't actually print the Z @notdict: bcs @notupper - adc #38 ; 27-52 are A-Z + adc #38 ; 27-51 are A-Y bne @printit @notupper: @@ -175,14 +229,17 @@ _print_msg: ldy ysave @noprint: lda #0 - sta outbyte + sta outsnac ldx #6 dec bitcount beq @nextbyte bne @bitloop dict_lookup: - ; dictionary lookup time. save our state on the stack + ; dictionary lookup time. save our state on the stack. note that + ; using the stack means dict entries could potentially contain + ; dictionary escapes. each level would eat 7 bytes of stack, so be + ; careful (the current dictionary doesn't do this at all) tya pha lda inbyte @@ -194,17 +251,16 @@ dict_lookup: lda bitcount pha - ; recursive call - ldx outbyte - lda dict_offsets,x + ldx outsnac ; get the start address of the dictionary entry into AX + lda dict_offsets,x ; this is why the dictionary can't be <255 bytes total clc - adc #<dict00 - sta dict_escape ; temp usage + adc #<dict00 ; calculate low byte from base + offset + sta dict_escape ; temp usage, will be overwritten after _print_msg lda #>dict00 - adc #0 - tax - lda dict_escape - jsr _print_msg + adc #0 ; calculate hi byte + tax ; hi byte in X + lda dict_escape ; lo byte in A + jsr _print_msg ; recursive call, print the dictionary entry ; restore old state lda #0 @@ -219,7 +275,8 @@ dict_lookup: sta inbyte pla tay - rts + + rts ; print rest of original message decodersize = * - _print_msg |