; Prog8 definitions for floating point handling on the Commodore-64 %option enable_floats %import floats_functions floats { ; ---- this block contains C-64 floating point related functions ---- %option no_symbol_prefixing const float PI = 3.141592653589793 const float TWOPI = 6.283185307179586 ; ---- C64 basic and kernal ROM float constants and functions ---- ; note: the fac1 and fac2 are working registers and take 6 bytes each, ; floats in memory (and rom) are stored in 5-byte MFLPT packed format. ; note: fac1/2 might get clobbered even if not mentioned in the function's name. ; note: for subtraction and division, the left operand is in fac2, the right operand in fac1. romsub $bba2 = MOVFM(uword mflpt @ AY) clobbers(A,Y) ; load mflpt value from memory in A/Y into fac1 romsub $bba6 = FREADMEM() clobbers(A,Y) ; load mflpt value from memory in $22/$23 into fac1 romsub $ba8c = CONUPK(uword mflpt @ AY) clobbers(A,Y) ; load mflpt value from memory in A/Y into fac2 romsub $ba90 = FAREADMEM() clobbers(A,Y) ; load mflpt value from memory in $22/$23 into fac2 romsub $bbfc = MOVFA() clobbers(A,X) ; copy fac2 to fac1 romsub $bc0c = MOVAF() clobbers(A,X) ; copy fac1 to fac2 (rounded) romsub $bc0f = MOVEF() clobbers(A,X) ; copy fac1 to fac2 romsub $bbd4 = MOVMF(uword mflpt @ XY) clobbers(A,Y) ; store fac1 to memory X/Y as 5-byte mflpt ; fac1 -> unsigned word in Y/A (might throw ILLEGAL QUANTITY) (result also in $14/15) ; (tip: use floats.GETADRAY to get A/Y output; lo/hi switched to normal little endian order) romsub $b7f7 = GETADR() clobbers(X) -> ubyte @ Y, ubyte @ A romsub $bc9b = QINT() clobbers(A,X,Y) ; fac1 -> 4-byte signed integer in 98-101 ($62-$65), with the MSB FIRST. romsub $b1bf = AYINT() clobbers(A,X,Y) ; fac1-> signed word in 100-101 ($64-$65) MSB FIRST. (might throw ILLEGAL QUANTITY) ; GIVAYF: signed word in Y/A (note different lsb/msb order) -> float in fac1 ; (tip: use floats.GIVAYFAY to use A/Y input; lo/hi switched to normal order) ; there is also floats.GIVUAYFAY - unsigned word in A/Y (lo/hi) to fac1 ; there is also floats.FREADS32 that reads from 98-101 ($62-$65) MSB FIRST ; there is also floats.FREADUS32 that reads from 98-101 ($62-$65) MSB FIRST ; there is also floats.FREADS24AXY that reads signed int24 into fac1 from A/X/Y (lo/mid/hi bytes) romsub $b391 = GIVAYF(ubyte lo @ Y, ubyte hi @ A) clobbers(A,X,Y) romsub $b3a2 = FREADUY(ubyte value @ Y) clobbers(A,X,Y) ; 8 bit unsigned Y -> float in fac1 romsub $bc3c = FREADSA(byte value @ A) clobbers(A,X,Y) ; 8 bit signed A -> float in fac1 romsub $b7b5 = FREADSTR(ubyte length @ A) clobbers(A,X,Y) ; str -> fac1, $22/23 must point to string, A=string length. Also see parse_f() romsub $aabc = FPRINTLN() clobbers(A,X,Y) ; print string of fac1, on one line (= with newline) destroys fac1. (consider FOUT + STROUT as well) romsub $bddd = FOUT() clobbers(X) -> uword @ AY ; fac1 -> string, address returned in AY ($0100) romsub $b849 = FADDH() clobbers(A,X,Y) ; fac1 += 0.5, for rounding- call this before INT romsub $bae2 = MUL10() clobbers(A,X,Y) ; fac1 *= 10 romsub $bafe = DIV10() clobbers(A,X,Y) ; fac1 /= 10 , CAUTION: result is always positive! romsub $bc5b = FCOMP(uword mflpt @ AY) clobbers(X,Y) -> ubyte @ A ; A = compare fac1 to mflpt in A/Y, 0=equal 1=fac1 is greater, 255=fac1 is less than romsub $b86a = FADDT() clobbers(A,X,Y) ; fac1 += fac2 romsub $b867 = FADD(uword mflpt @ AY) clobbers(A,X,Y) ; fac1 += mflpt value from A/Y romsub $b853 = FSUBT() clobbers(A,X,Y) ; fac1 = fac2-fac1 mind the order of the operands romsub $b850 = FSUB(uword mflpt @ AY) clobbers(A,X,Y) ; fac1 = mflpt from A/Y - fac1 romsub $ba2b = FMULTT() clobbers(A,X,Y) ; fac1 *= fac2 romsub $ba28 = FMULT(uword mflpt @ AY) clobbers(A,X,Y) ; fac1 *= mflpt value from A/Y romsub $bb12 = FDIVT() clobbers(A,X,Y) ; fac1 = fac2/fac1 mind the order of the operands romsub $bb0f = FDIV(uword mflpt @ AY) clobbers(A,X,Y) ; fac1 = mflpt in A/Y / fac1 romsub $bf7b = FPWRT() clobbers(A,X,Y) ; fac1 = fac2 ** fac1 romsub $bf78 = FPWR(uword mflpt @ AY) clobbers(A,X,Y) ; fac1 = fac2 ** mflpt from A/Y romsub $bd7e = FINLOG(byte value @A) clobbers (A, X, Y) ; fac1 += signed byte in A romsub $aed4 = NOTOP() clobbers(A,X,Y) ; fac1 = NOT(fac1) romsub $bccc = INT() clobbers(A,X,Y) ; INT() truncates, use FADDH first to round instead of trunc romsub $b9ea = LOG() clobbers(A,X,Y) ; fac1 = LN(fac1) (natural log) romsub $bc39 = SGN() clobbers(A,X,Y) ; fac1 = SGN(fac1), result of SIGN (-1, 0 or 1) romsub $bc2b = SIGN() -> ubyte @ A ; SIGN(fac1) to A, $ff, $0, $1 for negative, zero, positive romsub $bc58 = ABS() ; fac1 = ABS(fac1) romsub $bf71 = SQR() clobbers(A,X,Y) ; fac1 = SQRT(fac1) romsub $bf74 = SQRA() clobbers(A,X,Y) ; fac1 = SQRT(fac2) romsub $bfed = EXP() clobbers(A,X,Y) ; fac1 = EXP(fac1) (e ** fac1) romsub $bfb4 = NEGOP() clobbers(A) ; switch the sign of fac1 (fac1 = -fac1) romsub $e097 = RND() clobbers(A,X,Y) ; fac1 = RND(fac1) float random number generator romsub $e264 = COS() clobbers(A,X,Y) ; fac1 = COS(fac1) romsub $e26b = SIN() clobbers(A,X,Y) ; fac1 = SIN(fac1) romsub $e2b4 = TAN() clobbers(A,X,Y) ; fac1 = TAN(fac1) romsub $e30e = ATN() clobbers(A,X,Y) ; fac1 = ATN(fac1) asmsub FREADS32() clobbers(A,X,Y) { ; ---- fac1 = signed int32 from $62-$65 big endian (MSB FIRST) %asm {{ lda $62 eor #$ff asl a lda #0 ldx #$a0 jmp $bc4f ; internal BASIC routine }} } asmsub FREADUS32 () clobbers(A,X,Y) { ; ---- fac1 = uint32 from $62-$65 big endian (MSB FIRST) %asm {{ sec lda #0 ldx #$a0 jmp $bc4f ; internal BASIC routine }} } asmsub FREADS24AXY (ubyte lo @ A, ubyte mid @ X, ubyte hi @ Y) clobbers(A,X,Y) { ; ---- fac1 = signed int24 (A/X/Y contain lo/mid/hi bytes) ; note: there is no FREADU24AXY (unsigned), use FREADUS32 instead. %asm {{ sty $62 stx $63 sta $64 lda $62 eor #$FF asl a lda #0 sta $65 ldx #$98 jmp $bc4f ; internal BASIC routine }} } asmsub GIVUAYFAY (uword value @ AY) clobbers(A,X,Y) { ; ---- unsigned 16 bit word in A/Y (lo/hi) to fac1 %asm {{ sty $62 sta $63 ldx #$90 sec jmp $bc49 ; internal BASIC routine }} } asmsub GIVAYFAY (uword value @ AY) clobbers(A,X,Y) { ; ---- signed 16 bit word in A/Y (lo/hi) to float in fac1 %asm {{ sta P8ZP_SCRATCH_REG tya ldy P8ZP_SCRATCH_REG jmp GIVAYF ; this uses the inverse order, Y/A }} } asmsub GETADRAY () clobbers(X) -> uword @ AY { ; ---- fac1 to unsigned word in A/Y %asm {{ jsr GETADR ; this uses the inverse order, Y/A sta P8ZP_SCRATCH_B1 tya ldy P8ZP_SCRATCH_B1 rts }} } &uword AYINT_facmo = $64 ; $64/$65 contain result of AYINT sub rndf() -> float { %asm {{ lda #1 jsr FREADSA jmp RND ; rng into fac1 }} } asmsub parse_f(str value @AY) -> float @FAC1 { %asm {{ sta $22 sty $23 jsr prog8_lib.strlen tya jmp FREADSTR }} } %asminclude "library:c64/floats.asm" %asminclude "library:c64/floats_funcs.asm" }