brooke/mandel-6502
1
0
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity

apply jamey's suggestion of skipping add for high byte muls

Browse source 
rather than saving 0 into the high bytes, then adding the high-byte
multiplication later, write it directly in place. this saves a few
cycles on every iteration, and it adds up nicely.

View 1 overview render times:
130XE: 10.050 ms/px - 4m56s
800XL: 10.906 ms/px - 5m21s
This commit is contained in:
Brooke Vibber 2025-01-04 10:53:51 -08:00
parent d157fe1306
commit 582ddf497f
2 changed files with 4 additions and 28 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

29
mandel.s
View file

@ -464,20 +464,6 @@ viewport_oy:
sta dest + 1 sta dest + 1
.endmacro .endmacro
; input: arg as u8
; input/output: dest as u16
; clobbers a, x
.macro sqr8_add16 dest, arg
ldx arg
clc
lda sqr_lobyte,x
adc dest
sta dest
lda sqr_hibyte,x
adc dest + 1
sta dest + 1
.endmacro
.segment "TABLES" .segment "TABLES"
; lookup table for top byte -> PORTB value for bank-switch ; lookup table for top byte -> PORTB value for bank-switch
.align 256 .align 256
@ -760,9 +746,8 @@ inner_loop:
; h1*h2*256*256 + h1*l2*256 + h2*l1*256 + l1*l2 ; h1*h2*256*256 + h1*l2*256 + h2*l1*256 + l1*l2
imul8 result, arg1, arg2, xe imul8 result, arg1, arg2, xe
lda #0
sta result + 2 imul8 result + 2, arg1 + 1, arg2 + 1, xe
sta result + 3
imul8 inter, arg1 + 1, arg2, xe imul8 inter, arg1 + 1, arg2, xe
add16 result + 1, result + 1, inter add16 result + 1, result + 1, inter
@ -772,9 +757,6 @@ inner_loop:
add16 result + 1, result + 1, inter add16 result + 1, result + 1, inter
add_carry result + 3 add_carry result + 3
imul8 inter, arg1 + 1, arg2 + 1, xe
add16 result + 2, result + 2, inter
; In case of negative inputs, adjust high word ; In case of negative inputs, adjust high word
; https://stackoverflow.com/a/28827013 ; https://stackoverflow.com/a/28827013
lda arg1 + 1 lda arg1 + 1
@ -807,9 +789,8 @@ arg2_pos:
; h*h*256*256 + h*l*256 + h*l*256 + l*l ; h*h*256*256 + h*l*256 + h*l*256 + l*l
sqr8 result, arg sqr8 result, arg
lda #0
sta result + 2 sqr8 result + 2, arg + 1
sta result + 3
imul8 inter, arg + 1, arg, xe imul8 inter, arg + 1, arg, xe
add16 result + 1, result + 1, inter add16 result + 1, result + 1, inter
@ -817,8 +798,6 @@ arg2_pos:
add16 result + 1, result + 1, inter add16 result + 1, result + 1, inter
add_carry result + 3 add_carry result + 3
sqr8_add16 result + 2, arg + 1
rts ; 6 cyc rts ; 6 cyc
.endscope .endscope
.endmacro .endmacro

3
todo.md
View file

@ -1,8 +1,5 @@
things to try: things to try:
* skip add on the top-byte multiply in sqr8/mul8
* should save a few cycles, suggestion by jamey
* patch the entire expanded-ram imul8xe on top of imul8 to avoid the 3-cycle thunk penalty :D * patch the entire expanded-ram imul8xe on top of imul8 to avoid the 3-cycle thunk penalty :D
* y-axis mirror optimization * y-axis mirror optimization