more multiplication/division work

This commit is contained in:
FeralChild64 committed 2020-02-29 12:16:54 +01:00
1 parent d762c49312
commit b346cc85c5
5 files changed
+75 -145

No files matched your search

-1
View File
@@ -7,7 +7,6 @@
* Graham - for his PAL/NTSC detection routine
* enthusi/Onslaught - for his example Turbo Tape 64 compatible loader
* djmips - for fast 8x8 multiplication routine
* Verz - for integer division routine
## for documentation
+1 -1
View File
@@ -199,7 +199,7 @@ Floating point mathematical routines - not official, but well known and broadly
| `$B947` | `inv_FAC1_mantissa` | NOT DONE | |
| `$B9EA` | `log_FAC1` | NOT DONE | |
| `$BA28` | `mul_MEM_FAC1` | DONE | |
| `$BA2B` | `mul_FAC2_FAC1` | DONE | |
| `$BA2B` | `mul_FAC2_FAC1` | PARTIAL | needs retest after rework |
| `$BA8C` | `mov_MEM_FAC2` | DONE | |
| `$BA90` | `get_FAC2_via_INDEX` | NOT DONE | |
| `$BAE2` | `mul10_FAC1` | DONE | |
+18 -97
View File
@@ -34,7 +34,7 @@ div_FAC2_FAC1:
// Handle special case - dividing 0 by non-zero value
lda FAC2_exponent
beq div_FAC2_FAC1_setexp_quit
beq_16 set_FAC1_zero
// Multiply signs
@@ -42,29 +42,32 @@ div_FAC2_FAC1:
eor FAC2_sign
sta FAC1_sign
// Subtract the exponents XXX what about bias?
// Subtract the exponents
lda #($80 + $08) // correction for bias and shifted mantissa
sta RESHO+0
lda #$00
sta RESHO+1
lda FAC2_exponent
jsr muldiv_RESHO_01_add_A
lda RESHO+0
sec
sbc FAC1_exponent
bcs !+
// Result too low, set 0 and quit
lda #$00
div_FAC2_FAC1_setexp_quit:
sta FAC1_exponent
rts
lda RESHO+1
sbc #$00
bcc_16 set_FAC1_zero // result too low, set 0 and quit
!:
adc #$07 // add 8, as the carry is set here
bcs_16 set_FAC1_max // overflow
sta FAC1_exponent
lda RESHO+1
bcc_16 set_FAC1_max // overflow
// Divide the mantissas
jsr muldiv_RESHO_set_0
// XXX
@@ -74,87 +77,5 @@ div_FAC2_FAC1_setexp_quit:
// XXX adapt, move this to a separate file
// Unsigned integer 32-bit division, based on code by Verz
// divident: FAC2_mantissa, divisor: FAC1_mantissa, result: FAC2_mantissa, remainder: RESHO
// temporary storage: INDEX+1, INDEX+2, INDEX+3
//
// https://codebase64.org/doku.php?id=base:24bit_division_24-bit_result
// XXX test this, optimize (for temporary storage use .Y, .Z and RESHO+4)
div_mantissas:
// Preset remainder to 0
// XXX should be common with multiplication
lda #$00
sta RESHO+0
sta RESHO+1
sta RESHO+2
sta RESHO+3
ldx #$32 // repeat for each bit
// FALLTROUGH
div_mantissas_loop:
// Multiply divident by 2, msb -> Carry
asl FAC2_mantissa+3
rol FAC2_mantissa+2
rol FAC2_mantissa+1
rol FAC2_mantissa+0
// Multiply remainder by 2, lsb <- Carry
rol RESHO+3
rol RESHO+2
rol RESHO+1
rol RESHO+0
// Subtract divisor to see if it fits in
lda RESHO+3
sec
sbc FAC1_mantissa+2
sta INDEX+3
lda RESHO+2
sbc FAC1_mantissa+2
sta INDEX+2
lda RESHO+1
sbc FAC1_mantissa+1
sta INDEX+1
lda RESHO+0
sbc FAC1_mantissa+0
bcc div_mantissas_skip // if Carry is not set then divisor did not fit in yet
sta RESHO+0 // else save substraction result as new remainder,
lda INDEX+1
sta RESHO+1
lda INDEX+2
sta RESHO+2
lda INDEX+3
sta RESHO+3
inc FAC2_mantissa+3 // and INCrement result cause divisor fit in 1 times
// FALLTROUGH
div_mantissas_skip:
dex
bne div_mantissas_loop
rts
// XXX use 8-bit divide from here: https://codebase64.org/doku.php?id=base:8bit_divide_8bit_product
+27 -46
View File
@@ -17,6 +17,11 @@
// - https://codebase64.org/doku.php?id=base:kernal_floating_point_mathematics
//
set_FAC1_zero:
lda #$00
sta FAC1_exponent
mul_FAC2_FAC1_done:
rts
@@ -39,59 +44,35 @@ mul_FAC2_FAC1:
eor FAC2_sign
sta FAC1_sign
// Check if exponents can be added without overflow/underflow
lda FAC1_exponent
and #$80
eor FAC2_exponent
bmi mul_FAC2_FAC1_add_exponents // exponent signs are opposite, we can add them safely
// Both exponents are either negative (below $80), or both are non-negative ($80 or above),
// try to add them
lda FAC1_exponent
clc
adc FAC2_exponent
bpl mul_FAC2_FAC1_sub_exp_bias // branch if we can add values
// We cannot add them, result would over/underflow; set result as
// either 0 or maximum possible float
lda FAC1_exponent
bmi_16 set_FAC1_max
// Add the exponents (subtract the bias) XXX subroutines should be common with DIV
lda #$00
sta FAC1_exponent
rts
mul_FAC2_FAC1_add_exponents:
// Add the exponents
lda FAC1_exponent
clc
adc FAC2_exponent
// FALLTROUGH
mul_FAC2_FAC1_sub_exp_bias:
sec
sbc #$80 // bias
sta FAC1_exponent
// Multiply the mantissas - first clear the RESHO temporary area
jsr muldiv_RESHO_set_0
lda #$00
lda FAC1_exponent
jsr muldiv_RESHO_01_add_A
lda FAC2_exponent
jsr muldiv_RESHO_01_add_A
lda RESHO+0
sbc #($80 - $08) // we need to correct double BIAS and shifted mantissa
sta RESHO+0
bcs !+
lda RESHO+1
sbc #$00
bcs set_FAC1_zero
sta RESHO+1
sta RESHO+2
sta RESHO+3
sta RESHO+4
!:
lda RESHO+1
bne_16 set_FAC1_max // overflow
lda RESHO+0
sta FAC1_exponent
// Multiply the mantissas
jsr muldiv_RESHO_set_0
// XXX maybe we should preserve .X on stack? this will not cost us too much
lda FACOV
+29
View File
@@ -0,0 +1,29 @@
// #LAYOUT# STD * #TAKE
// #LAYOUT# * BASIC_0 #TAKE
// #LAYOUT# * * #IGNORE
//
// Math package - helper routines for RESHO manipulation
//
muldiv_RESHO_set_0:
lda #$00
sta RESHO+0
sta RESHO+1
sta RESHO+2
sta RESHO+3
sta RESHO+4
rts
muldiv_RESHO_01_add_A:
clc
adc RESHO+0
sta RESHO+0
bcc !+
inc RESHO+1
!:
rts