Mercurial > public > hwos_code
view src/math.asm @ 411:93b270b3503d
BUGFIX: Fix missing ccr setup menu (From 2.02)
author | heinrichsweikamp |
---|---|
date | Fri, 11 Mar 2016 10:11:07 +0100 |
parents | 653a3ab08062 |
children | ca4556fb60b9 |
line wrap: on
line source
;============================================================================= ; ; File math.asm ; ; Math subroutines ; ; Copyright (c) 2011, JD Gascuel, HeinrichsWeikamp, all right reserved. ;============================================================================= ; HISTORY ; 2011-08-03 : [mH] moving from OSTC code #include "hwos.inc" ; Mandatory header basic CODE ;============================================================================= global convert_time convert_time: ; converts hi:lo in minutes to hours (hi) and minutes (lo) movff lo,xA+0 ; divide by 60... movff hi,xA+1 ; movlw d'60' ; movwf xB+0 ; clrf xB+1 ; rcall div16x16 ; xA/xB=xC with xA as remainder movff xC+0,hi ; Hours movff xA+0,lo ; =remaining minutes (0.....59) return global div16 div16: ; divA=divA/2^divB (divB: 8Bit only!) bcf STATUS,C rrcf divA+1 rrcf divA+0 decfsz divB bra div16 return global sub16 sub16: ; sub_c = sub_a - sub_b (with signed values) bcf neg_flag movf sub_b+0, W ; Get Value to be subtracted subwf sub_a+0, W ; Do the High Byte movwf sub_c+0 movf sub_b+1, W ; Get the Value to be Subbed subwfb sub_a+1, W movwf sub_c+1 btfss STATUS,N ; Negativ result ? return ; NO: result positive done. bsf neg_flag ; MARK result negative comf sub_c+1 ; 16bit sign change. negf sub_c+0 btfsc STATUS,C ; Carry to propagate ? incf sub_c+1,F ; YES: do it. return global subU16 subU16: ; sub_c = sub_a - sub_b (with UNSIGNED values) bcf neg_flag movf sub_b+0, W ; Get Value to be subtracted subwf sub_a+0, W ; Do the High Byte movwf sub_c+0 movf sub_b+1, W ; Get the Value to be Subbed subwfb sub_a+1, W movwf sub_c+1 btfsc STATUS,C ; Borrow to propagate ? (B == /CARRY) return ; NO: result positive done. bsf neg_flag ; MARK result negative comf sub_c+1 ; 16bit sign change. negf sub_c+0 btfsc STATUS,C ; Carry to propagate ? incf sub_c+1,F ; YES: do it. return global mult16x16 mult16x16: ;xA*xB=xC clrf xC+2 ; Clear the High-Order Bits clrf xC+3 movf xA, w ; Do the "L" Multiplication first mulwf xB movf PRODL, w ; Save result movwf xC movf PRODH, w movwf xC+1 movf xA, w ; Do the "I" Multiplication mulwf xB+1 movf PRODL, w ; Save the Most Significant Byte First addwf xC+1, f movf PRODH, w addwfc xC+2, f ; Add to the Last Result movf xA+1, w ; Do the "O" Multiplication mulwf xB movf PRODL, w ; Add the Lower Byte Next addwf xC+1, f movf PRODH, w ; Add the High Byte First addwfc xC+2, f btfsc STATUS, C ; Add the Carry incf xC+3, f movf xA+1, w ; Do the "F" Multiplication mulwf xB+1 movf PRODL, w addwf xC+2, f movf PRODH, w addwfc xC+3, f return global div16x16 div16x16: ;xA/xB=xC with xA+0 as remainder ;uses divB as temp variable clrf xC+0 clrf xC+1 MOVF xB+0,W ; Check for zero IORWF xB+1,W ; BTFSC STATUS,Z ; Check for zero RETLW H'FF' ; return 0xFF if illegal MOVLW 1 ; Start count at 1 MOVWF divB ; Clear Count div16x16_1: BTFSC xB+1,7 ; High bit set ? bra div16x16_2 ; Yes then continue INCF divB,F ; Increment count bcf STATUS,C rlcf xB+0,F rlcf xB+1,F bra div16x16_1 div16x16_2: ; Shift result left bcf STATUS,C rlcf xC+0,F rlcf xC+1,F ; Reduce Divisor MOVF xB,W ; Get low byte of subtrahend SUBWF xA,F ; Subtract DST(low) - SRC(low) MOVF xB+1,W ; Now get high byte of subtrahend BTFSS STATUS,C ; If there was a borrow, rather than INCF xB+1,W ; decrement high byte of dst we inc src SUBWF xA+1,F ; Subtract the high byte and we're done BTFSC STATUS, C ; Did it reduce? bra div16x16_3 ; No, so it was less than movf xB+0,W ; Reverse subtraction addwf xA+0,F movf xB+1,W addwfc xA+1,F bra div16x16_4 ; Continue the process div16x16_3: BSF xC+0,0 ; Yes it did, this gets a 1 bit div16x16_4: DECF divB,F ; Decrement N_COUNT BTFSC STATUS,Z ; If its not zero then continue return bcf STATUS,C rrcf xB+1,F rrcf xB+0,F bra div16x16_2 ; Next bit. global div32x16 div32x16: ; xC:4 / xB:2 = xC+3:xC+2 with xC+1:xC+0 as remainder ; Setup movlw .32 ; setup shift counter movwf divB movff xC+3,xA+1 ; move ACCb to ACCf movff xC+2,xA+0 movff xC+1,sub_a+1 ; move ACCc to ACCe movff xC+0,sub_a+0 clrf xC+3 clrf xC+2 clrf xC+1 clrf xC+0 clrf sub_b+1 clrf sub_b+0 div32x16_2: bcf STATUS,C rlcf sub_a+0,F rlcf sub_a+1,F rlcf xA+0,F rlcf xA+1,F rlcf sub_b+0,F rlcf sub_b+1,F movf xB+1,W subwf sub_b+1,W ; check if a>d btfss STATUS,Z bra div32x16_3 movf xB+0,W subwf sub_b+0,W ; if msb equal then check lsb div32x16_3: btfss STATUS,C ; carry set if d>a bra div32x16_4 movf xB+0,W ; d-a into d subwf sub_b+0,F btfss STATUS,C decf sub_b+1,F movf xB+1,W subwf sub_b+1,F bsf STATUS,C ; shift a 1 into b (result) div32x16_4: rlcf xC+0,F rlcf xC+1,F rlcf xC+2,F rlcf xC+3,F decfsz divB,F ; loop until all bits checked bra div32x16_2 return ;;============================================================================= ;; u16 * u16 --> 32bit multiply (xA * xB --> xC) ;; Used in interupt service routines, to compute temperature and pressure. ;; ; global isr_mult16x16 ;isr_mult16x16: ; clrf isr_xC+2 ; Clear the High-Order Bits ; clrf isr_xC+3 ; movf isr_xA, w ; Do the "L" Multiplication first ; mulwf isr_xB ; movf PRODL, w ; Save result ; movwf isr_xC+0 ; movf PRODH, w ; movwf isr_xC+1 ; movf isr_xA+0, w ; Do the "I" Multiplication ; mulwf isr_xB+1 ; movf PRODL, w ; Save the Most Significant Byte First ; addwf isr_xC+1, f ; movf PRODH, w ; addwfc isr_xC+2, f ; Add to the Last Result ; movf isr_xA+1, w ; Do the "O" Multiplication ; mulwf isr_xB ; movf PRODL, w ; Add the Lower Byte Next ; addwf isr_xC+1, f ; movf PRODH, w ; Add the High Byte First ; addwfc isr_xC+2, f ; btfsc STATUS, C ; Add the Carry ; incf isr_xC+3, f ; movf isr_xA+1, w ; Do the "F" Multiplication ; mulwf isr_xB+1 ; movf PRODL, w ; addwf isr_xC+2, f ; movf PRODH, w ; addwfc isr_xC+3, f ; return ;============================================================================= ; 24bit shift, repeted WREG times. ; Because we shift less than 8bits, and keep only C[2:1], we don't care what ; bit is inserted... ; global isr_shift_C31 isr_shift_C31: rrcf isr_xC+3,F ; Shift the three bytes... rrcf isr_xC+2,F rrcf isr_xC+1,F decfsz WREG bra isr_shift_C31 return ;============================================================================= ; s16 * s16 --> 32bit multiply (xA * xB --> xC) ; Signed multiplication. ; Code from... the Pic18F documentation ;-) global isr_unsigned_mult16x16 isr_unsigned_mult16x16: MOVF isr_xA+0, W ; Lowest is simply a[0] * b[0] MULWF isr_xB+0 MOVFF PRODL, isr_xC+0 MOVFF PRODH, isr_xC+1 ; MOVF isr_xA+1, W ; And highest a[1] * b[1] MULWF isr_xB+1 MOVFF PRODL, isr_xC+2 MOVFF PRODH, isr_xC+3 ; MOVF isr_xA+0, W ; Intermediates do propagate: MULWF isr_xB+1 MOVF PRODL, W ADDWF isr_xC+1, F ; Add cross products MOVF PRODH, W ADDWFC isr_xC+2, F ; with propagated carry CLRF WREG ADDWFC isr_xC+3, F ; on the three bytes. ; MOVF isr_xA+1, W ; And the second one, similarly. MULWF isr_xB+0 MOVF PRODL, W ADDWF isr_xC+1, F ; Add cross products MOVF PRODH, W ADDWFC isr_xC+2, F CLRF WREG ADDWFC isr_xC+3, F return global isr_signed_mult16x16 isr_signed_mult16x16: rcall isr_unsigned_mult16x16 ; Manage sign extension of operand B BTFSS isr_xB+1,7 ; Is B negatif ? BRA isr_signed_mult_checkA ; No: check ARG1 MOVF isr_xA+0, W ; Yes: add -65536 * A SUBWF isr_xC+2, F MOVF isr_xA+1, W SUBWFB isr_xC+3, F ; And of operand A isr_signed_mult_checkA BTFSS isr_xA+1, 7 ; Is A negatif ? RETURN ; No: done MOVF isr_xB+0, W SUBWF isr_xC+2, F MOVF isr_xB+1, W SUBWFB isr_xC+3, F RETURN END