Gold Fish 2

home *** CD-ROM | disk | FTP | other *** search

/ Gold Fish 2 / goldfish_vol2_cd1.bin / files / dev / cross / math68hc11 / div48.a11 < prev next >

Wrap

Text File | 1988-03-23 | 14KB | 417 lines

* Demonstration program for 24 bit multiplication with 48 bit result * and 24 bit division of a 48 bit dividend. * * The multiplication routine is typical for 6803 class machines. * * The division routine is unusual in that it uses the divide instruction * of the 68HC11 to find a trial divisor, proceeding by bytes until the * division is complete. A divide takes about 800us. The code is * lengthy -- size was traded for speed. * * If you find an error or you improve these routines by making them * faster and shorter or if you have an approach which is faster, * please contact me. * * John Moran (203)-575-3016 * Bristol Babcock Inc. * 1100 Buckingham St. * Watertown, Conn. 06795 * * * ----------------------------------------------------------------- * Definition of RAM use, assumes M68HC11EVB Evaluation Board ORG $DD00 ; Start of RAM. CNTR RMB 1 ; Counter for mul/div DIVPTR RMB 2 ; Ptr to MS byte of divisor DIVCNT RMB 1 ; Number of bytes in the divisor DVDCNT RMB 1 ; Number of bytes in the dividend QPTR RMB 2 ; Ptr to current byte of quotient QCNT RMB 1 ; Number of bytes remaining to be found in quotient TRIALQ RMB 1 ; Current quotient trial value ORG $DD50 ; Temp storage for multiply A24 RMB 3 ; Argument for multiple precision mul B24 RMB 3 ; Argument for multiple precision mul A48 RMB 6 ; Product from multiple precision mul ORG $DD60 ; Temp storage for divide D24 RMB 3 ; Divisor Q24 RMB 3 ; Quotient - Must immediately precede dividend D48 RMB 6 ; Dividend ORG $DD70 ; Temp storage for divide TEMP RMB 2 ; Work space for divide * * * ----------------------------------------------------------------- * BUFFALO's terminal I/O Hooks OUT1BYT EQU $FFBB OUT1BSP EQU $FFBE OUTSTRG EQU $FFC7 OUTCRLF EQU $FFC4 * * ================================================================= * Code space - at C000-DFFF for debug. ORG $C000 * - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - * Test routine entries: BRA DIVTST ; Use output of last multiply as dividend BRA MUL24 BRA DIV24J DIV24J: JMP DIV24 ; Extend branch range * - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - * Calculate the 48 bit unsigned product of two 24 bit unsigned numbers. * Place the multiplicand in A24, the multiplier in B24, then JSR MUL24. * Result is returned in A48. Put smaller value in B24 when given a * choice. * Note: A48 must immediately follow B24 in memory !!!! * B24 may overlap A48 to save space if desired - some changes req'd. * The approach used is similar to multiplication with pencil and paper, * where bytes are used as if they were digits. It may be extended to * any reasonable number of bytes quite easily. MUL24: PSHB ; Preserve regs PSHA PSHX LDX #B24 ; Address of args CLR 3,X ; Clear A48 CLR 4,X CLR 5,X CLR 6,X CLR 7,X CLR 8,X LDAA #3 ; # bytes in multiplier STAA CNTR OPT c MULOOP: LDAA 2,X BEQ MULUP2 ; Skip if 0 multiplier LDAB A24+2 ; Least signifigant byte of multiplicand MUL ADDD 7,X STD 7,X BCC MUL1 INC 6,X ; Propagate the carry BNE MUL1 INC 5,X MUL1: LDAA 2,X LDAB A24+1 MUL ADDD 6,X STD 6,X BCC MUL2 INC 5,X ; Propagate Carry MUL2: LDAA 2,X LDAB A24 ; Most signifigant byte MUL ADDD 5,X STD 5,X ; No carry possible on hi end MULUP2: DEX ; Move left 1 byte in multiplier & result DEC CNTR BNE MULOOP PULX ; Restore regs PULA PULB RTS OPT noc * * * This is a test routine for MUL24 and DIV24. * It calculates the product of two 24 bit numbers and then * divides this product by each of the original factors, checking * to ensure that the result is the other factor. It halts if an * error is found. You can insert breakpoints if an error is found * and then start again at C000 (Don't reload!!!) and it will * repeat the offending calculation. * * Takes about 2ms for the multiply and 2 divides. * * After each successful test, the factors are adjusted by adding * a prime number to select the next numbers to test. * * I let this test run for 9 days to check these routines. * DIVTST: JSR MUL24 ; CALCULATE PRODUCT LDD A24 STD D24 LDAB A24+2 STAB D24+2 JSR DIVPROD LDAB B24 CMPB Q24 BNE FAIL ; QUIT IF ERROR LDAB B24+1 CMPB Q24+1 BNE FAIL LDAB B24+2 CMPB Q24+2 BNE FAIL LDD B24 ; CHECK THE MULTIPLIER TOO STD D24 LDAB B24+2 STAB D24+2 JSR DIVPROD LDAB A24 CMPB Q24 BNE FAIL ; QUIT IF ERROR LDAB A24+1 CMPB Q24+1 BNE FAIL LDAB A24+2 CMPB Q24+2 BNE FAIL * Looks OK, calculate next numbers to use as factors BUMPA: LDD #593 ; Prime number ADDD A24+1 STD A24+1 BCC BUMPB INC A24 ; Propagate carry BEQ BUMPA ; Avoid 0 BUMPB: LDX #B24 INC 2,X ; 1 is prime BNE BMPDUN JSR PRINT ; Show that we're still OK INC 1,X BNE BMPDUN INC 0,X BEQ BUMPB ; Avoid 0 BMPDUN: JMP DIVTST ; LOOP FOREVER * Quotient didn't match one of the factors FAIL: BSR PRINT LDX #FAILSTR JSR OUTSTRG ; Announce the failure DIE: NOP ; Simulate a halt JMP DIE PRINT: PSHX LDX #A24 ; PRINT THE LAST ITEMS MULTIPLIED JSR OUT1BYT JSR OUT1BYT JSR OUT1BSP JSR OUT1BYT JSR OUT1BYT JSR OUT1BSP JSR OUTCRLF PULX RTS FAILSTR FCC ' FAIL' FCB 04 * DIVPROD: LDX #A48 ; Copy product to dividend LDAA #6 ; 6 bytes = 48 bits DIVTLP: LDAB 0,X STAB $10,X INX DECA BNE DIVTLP * ; Drop into divide subroutine * * * Calculate the 24 bit unsigned quotient resulting from division of a * 48 bit dividend by a 24 bit divisor. Place the divisor in D24 and the * dividend in D48. The result will be returned in Q24. * The approach is similar to long division by hand, but here the * IDIV instruction is used to form the trial quotient. Care is taken * to ensure that the trial quotient is always <= the required quotient. * The division routine can be modified to handle any reasonable number * of bytes, but it is more challenging than extending the multiplication * routine. * Typical operation is two passes through the loop per byte in the * dividend. DIV24: PSHB PSHA PSHX CLR Q24 ; Clear quotient to 0 CLR Q24+1 CLR Q24+2 LDAB #6 ; Size of dividend, max LDY #D48 ; Count bytes in the dividend DIV0: TST 0,Y BNE DIV1 DECB BEQ DIVER1 ; Done if dividend = 0, Return 0 INY BRA DIV0 DIVER1: JMP DIVBY0 ; Extend branch range * Find size of divisor and dividend & thus the hi byte of quotient. * Also do gross checks to avoid dividing by 0, etc. DIV1: STAB DVDCNT ; Save # bytes in dividend LDAA #$FF NEGB ADDD #D48+5 ; + Addr of LS Byte-1 in dividend XGDY ; Y points to MSB of dividend LDA #3 ; Count bytes in the divisor LDX #D24 DIV1CT: TST 0,X BNE DIV2 DECA BEQ DIVER1 ; Quit if divide by 0, Return 0 INX BRA DIV1CT DIV2: STX DIVPTR ; Ptr to MSB of divisor STA DIVCNT ; Number of bytes in divisor SUBA DVDCNT BGT DIVER1 ; Quit if divisor > dividend, Return 0 LDAB 1,Y ; Get hi byte of dividend CMPB 0,X ; IF hi byte of dividend > hi byte of divisor BLO DIV3 ; THEN one more byte needed in the quotient DECA DEY ; and in Dividend DIV3: TAB NEGA ; Make it positive STAA QCNT ; # Bytes in quotient CMPA #4 ; ??? MAKE THIS PREVENT OVERFLOW ??? BGT DIVER1 ; Quit if quotient too big, return 0 LDAA #$FF ; Make it a negative 16 bit value ADDD #Q24+2 ; Ptr to hi byte of quotient STD QPTR * Decide whether to use 1 or 2 bytes as trial divisor DIVLUP: LDX DIVPTR ; Get ptr to MSB of divisor LDAA DIVCNT ; Get # bytes in divisor DECA BEQ DIV1BY ; Only 1 byte in divisor STA CNTR ; Temp, # bytes in divisor -1 LDD 1,Y ; Get hi word of dividend in B CPD 0,X ; IF hi word of dividend < hi word of divisor BLO DIVBYB ; THEN use only MSB of divisor BHI DIVW2B ; Divisor < Dividend & 2+ bytes in divisor LDA CNTR DECA BEQ DIVWRD ; BR if divisor = MSB's of dividend LDAB 3,Y ; Get next byte of dividend in B CMPB 2,X ; IF next byte of dividend < this byte of divisor BLO DIVBYB ; THEN use only MSB of divisor, rounded up LDAB #1 ; ELSE Trial Quotient = 1 BRA DIVSTO DIVBYB: JMP DIVBYT ; Extend branch range DIV1BY: JMP DIV1BYT * Trial divisor is 2 bytes. Round up if required. DIVW2B: LDA CNTR DECA ; Make A = 0 if exactly 2 bytes in divisor DIVWRD: LDX 0,X ; Get MSW of divisor TSTA BEQ DIVWNR ; BR if exactly 2 bytes in divisor INX ; Round divisor up DIVWNR: LDD 1,Y ; Get MSW of dividend IDIV XGDX ; Get trial quotient into B * Multiply divisor by trial quotient and subtract from dividend DIVSTO: STAB TRIALQ ; Save trial quotient LDX QPTR ADDB 0,X ; Add Trial Q to current Q STAB 0,X LDAB DIVCNT STAB CNTR LDX DIVPTR LDAA 0,X ; Get hi byte of divisor LDAB TRIALQ ; Get trial quotient byte MUL STD TEMP ; Seems crude, but it works... LDD 0,Y SUBD TEMP ; No borrow possible from hi byte STD 0,Y DEC CNTR BEQ DIVLND ; BR if divisor is 1 byte LDX DIVPTR LDAA 1,X ; Get next byte of divisor LDAB TRIALQ MUL STD TEMP LDD 1,Y SUBD TEMP STD 1,Y BCC *+5 DEC 0,Y ; Propagate borrow DEC CNTR BEQ DIVLND ; BR if divisor is 2 bytes LDX DIVPTR LDAA 2,X LDAB TRIALQ MUL STD TEMP LDD 2,Y SUBD TEMP STD 2,Y BCC DIVLND LDX 0,Y ; Propagate borrow DEX STX 0,Y DIVLND: LDD 0,Y BEQ DIVDN0 TSTA BEQ DIVLN2 ; Almost always branches, 4000:1 or so LDX DIVPTR ; Ptr to MSB of divisor BRA DIVBYR ; Last divide by byte was too coarse, correct it DIVDN0: INY ; Hi word of dividend=0, move 1 position right DEC QCNT BLT DIVDUN INC QPTR+1 DIVLN2: JMP DIVLUP ; Loop to do next byte * Trial divisor is hi byte of divisor. Always round it up. DIVBYT: DEC QCNT ; Divisor > dividend, move 1 position right BLT DIVDUN ; BR if done INC QPTR+1 INY DIVBYR: LDAB 0,X ; Get MSB of divisor CLRA XGDX INX ; Round divisor up LDD 0,Y ; Dividend, current hi 16 bits BEQ DIVDN0 ; Shortcut on 0 dividend ?? Does this help?? IDIV ; MSW of dividend / MSB of divisor = Trial Q XGDX ; Get quotient in D TSTA BEQ DIVBT2 LDAB #$F0 ; Overflow on this trial, force max DIVBT2: JMP DIVSTO * Handle single byte divisor specially DIV1BYT: LDAB 0,X ; Get divisor, single byte CLRA XGDX DIV1SKP: CPX 0,Y BLS DIV1BGO DEC QCNT ; Divisor > Dividend, move 1 position right BLT DIVDUN ; BR if done INC QPTR+1 INY BRA DIV1SKP DIV1BGO: LDD 0,Y ; Dividend, current hi 16 bits IDIV ; MSW of dividend / divisor = Trial Q XGDX ; Get quotient in D JMP DIVSTO DIVDUN: DIVBY0: PULX ; Restore regs PULA PULB RTS END